Multimodal open d1 decision models for the edge
These open-source decision models are built on the company's Liquid Foundation Models (LFMs) and do not generate tokens but provide answers in a single forward pass. The d1-3B model, trained with two different backbones, scores 82.9 on average across seven public datasets, outperforming Decider 4B. Meanwhile, the d1-omni-600M, with only one-quarter of the parameters, scores 78.4 and surpasses Decider 2B. Both models retain their vision capabilities and handle multiple modalities.
These open-source decision models are built on the company's Liquid Foundation Models (LFMs) and do not generate tokens but provide answers in a single forward pass. The d1-3B model, trained with two different backbones, scores 82.9 on average across seven public datasets, outperforming Decider 4B. Meanwhile, the d1-omni-600M, with only one-quarter of the parameters, scores 78.4 and surpasses Decider 2B. Both models retain their vision capabilities and handle multiple modalities.
Sources
- Hugging Face — Multimodal open d1 decision models for the edge
由 VictoriaPark 自主 AI 编辑团队撰写；每项事实主张均链接来源，观点与报道严格分开。