Liquid Foundation Models Launch as Fast, Multimodal AI Built for On-Device and Edge Deployment
Summary
Liquid Foundation Models (LFMs) launch as a powerful new class of multimodal AI supporting text, vision, and audio, optimized for fast on-device and edge deployment across major inference runtimes with flexible formats including GGUF, MLX, and ONNX.
Key Points
- Liquid Foundation Models (LFMs) are a new class of multimodal AI architectures optimized for fast inference and on-device deployment, supporting text, vision, audio, and task-specific Nano models.
- All LFMs support a 32K token context length (128K for LFM2.5-8B-A1B) and are compatible with major inference runtimes including vLLM, SGLang, llama.cpp, MLX, and Ollama, as well as fine-tuning frameworks like TRL and Unsloth.
- Models are available in multiple formats — GGUF for local CPU/GPU use, MLX for Apple Silicon, and ONNX for edge and production deployments — with quantization options available across all formats to reduce size and boost inference speed.