Liquid AI Release LFM2-VL: Ultra-fast open visual language model designed for low latency and device-aware deployments

Liquid AI has been officially released LFM2-VLa new visual foundation model, optimized for low latency, device deployment. With two efficient variants –LFM2-VL-450M and LFM2-VL-1.6B– This launch marks the bringing multimodal AI to smartphones, laptops, wearables and embedded systems without compromising speed or accuracy.

Unprecedented speed and efficiency

LFM2-VL models have been designed for delivery Up to 2× Faster GPU Reasoning Competitive benchmark performance is maintained on tasks such as image description, visual question answers, and multimodal reasoning compared to existing vision models. The 450m parameter variant is tailored to highly resource-constrained environments, while the 1.6B parameter version has greater functionality while still maintaining a lightweight and lightweight for single-GPU or high-end mobile use.

Technological innovation

• Modular architecture: LFM2-VL combines the language model backbone (LFM2-1.2B or LFM2-350M), Siglip2 Naflex visual encoder (400m or 86m parameters) and has a “Pixel Unshuffle” technology that dynamically reduces the “Pixel Unshuffle” technology of image token Counts.
• Local resolution processing: Images are in their Local resolution up to 512×512 pixels No distortion to the upgrade. Larger images are divided into non-overlapping 512×512 patches, retaining details and aspect ratios. The 1.6B model also encodes a zoomed-out thumbnail of the complete image to understand the global context.
• Flexible reasoning: Users can Trade-offs on adjusting speed quality when reasoning Adapt device functionality and application requirements in real time by adjusting the maximum image token and patch count.
• train: These models were first pre-trained on the LFM2 main chain, then fuse visual and language abilities by gradually adjusting the text to image data ratio, and ultimately perform image understanding of approximately 100 billion multimodal tokens.

Benchmark performance

Provided by LFM2-VL Competition results On public benchmarks such as Realworldqa, MM-Ifeval, and Ocrbench, it is comparable to larger numbers such as Internvl3 and Smolvlm2, but there is one Smaller memory footprint And faster processing – make it perfect for edge and mobile applications.

Both models are of size Open weight and downloadable Under the hugging face Apache 2.0-based licenseallowing free use of company research and commercial use. Larger businesses must contact Liquid AI to obtain a commercial license. These models will seamlessly integrate with embracing surface transformers and support quantization to further improve edge hardware efficiency.

Use cases and integration

LFM2-VL is designed for developers and businesses seeking deployment Fast, accurate, efficient multimodal AI Directly on the device – Reduce cloud dependency and enable new applications in robots, IoT, smart cameras, mobile assistants and more. Sample applications include live image subtitles, visual searches, and interactive multimodal chatbots.

getting Started

• download: Both models are now available in the Liquid AI Hug Face Collection.
• running: For example, inference code is provided for platforms such as Llama.cpp, supporting various quantization levels for optimal performance on different hardware.
• custom made: This architecture supports integration with Liquid AI’s LEAP platform for further customization and multi-platform edge deployment.

AnywayLiquid AI’s LFM2-VL is effective on the edge, and the open visual language model sets a new standard. With native resolution support, tradeoffs on adjustable speed quality and focus on real-world deployments, it enables developers to build next-generation AI-powered applications on any device.

Check Technical details and Model embracing face. Check out ours anytime Tutorials, codes and notebooks for github pages. Also, please stay tuned for us twitter And don’t forget to join us 100K+ ml reddit And subscribe Our newsletter.

Asif Razzaq is CEO of Marktechpost Media Inc. As a visionary entrepreneur and engineer, ASIF is committed to harnessing the potential of artificial intelligence to achieve social benefits. His recent effort is to launch Marktechpost, an artificial intelligence media platform that has an in-depth coverage of machine learning and deep learning news that can sound both technically, both through technical voices and be understood by a wide audience. The platform has over 2 million views per month, demonstrating its popularity among its audience.