Multimodal open d1 decision models for the edge
huggingface ·
LiquidAI has released two open-weight decision models, d1-3B and the experimental d1-omni-600M, designed for efficient edge computing. Unlike traditional generative models that produce tokens sequentially, these decision models provide answers in a single forward pass, ensuring significantly faster inference speeds. The d1-3B model supports text and image inputs, while d1-omni-600M handles multimodal inputs including text, images, and audio. Benchmarks indicate that d1-3B achieves the highest Decision Index score among models under 10 billion parameters, outperforming larger competitors like Decider 35B-A3B. In terms of speed, d1-3B can answer questions in as little as 8 milliseconds on high-end GPUs and under 50 milliseconds on NVIDIA Jetson edge devices. These models are built upon Liquid Foundation Models, with d1-3B derived from a decoder-only VLM and d1-omni-600M from a bidirectional encoder architecture. Both models are available for download on Hugging Face, allowing developers to integrate fast, structured decision-making capabilities into their applications.
huggingface