Llama app adds one-click local decision models: d3-flash tops Jev Index, d1-Omni-2B goes omni
huggingface · x · 2026-11-05
The llama.cpp team and Hugging Face's Llama desktop app now run "decision models" locally with a single click: feed the model a document plus typed questions and get a calibrated probability for every option, fast.
- d3-flash (vLLM-SR's Decision 3.0 family): #1 on the Jev Decision Index (98), reads text, JSON, images and video; 6.6 GB Q4KM, Apache-2.0.
- d1-Omni-2B (Liquid AI): an omni decision model handling text, images, audio and even spoken questions in just 2B parameters (1.85 GB Q80), laptop-friendly.
Target use cases include ticket routing, content moderation, and picking an agent's next step.
More from Infra
- Eon launches Yukon serverless data engine, arguing AI adoption is bottlenecked by data infra, not models — EXM7777 · 2026-11-05
- SK Hynix's Solidigm plans two New York NAND fabs in first US manufacturing push, up to $100B — Beth_Kindig · 2026-11-05
- Trainium's biggest hurdle isn't the chip — it's the CUDA ecosystem — DavidLinthicum · 2026-11-05
- Datacenter slowdown as bubble cure? Tech stocks slide as debate rages — oran_ge · 2026-11-05
- Nebius auto-tuner speeds RL training up to 5x on fixed GPU budgets — demian_ai · 2026-11-05
- After communities vote against data centers, OpenAI pledges to pay power bills — ross2000 · 2026-11-05