Liquid AI has open-sourced two of its d1 decision models, d1-3B and d1-omni-600M, making them available on Hugging Face. These models are designed to process various inputs like text, images, and audio to provide structured, typed answers to predefined questions, aiming to replace traditional LLM calls for tasks such as ticket routing and intent classification. The d1-3B model, built on LFM2.5-VL-3B, claims state-of-the-art performance for models under 10 billion parameters, with reported fast inference speeds on consumer hardware. The d1-omni-600M is an experimental multimodal model for edge devices, and the author provides a guide for running it on a CPU. AI
IMPACT Enables local processing of decision-making tasks, potentially reducing reliance on cloud-based LLM APIs for specific applications.
RANK_REASON Open-source release of specialized AI models from a company. [lever_c_demoted from research: ic=1 ai=1.0]
- Apple M5 Pro
- d1-3B
- d1-omni-600M
- Hugging Face
- LFM2.5-VL-3B
- Liquid AI
- llama.cpp
- NVIDIA Jetson Orin Nano 8GB
- OpenAI
- RTX 4090
- X
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
