JEV-9B passes 336K downloads with 0.5% parameter training and a System 1 decision block
AutoTrust's JEV-9B uses Blocks of Experts to train only 0.5% of parameters plus a calibrated System 1 decision block while keeping the original reasoning weights, and supports tool routing and confidence-gated agent workflows.
Thank you @HuggingModels for highlighting AutoTrust AI’s JEV-9B — now at 336K downloads! Our Blocks of Experts recipe adds a calibrated System 1 decision block by training just 0.5% of the backbone’s parameters, while preserving its original System 2 reasoning weights. One deployment for fast decisions and deeper reasoning. Build tool routers, classifiers, and confidence-gated agent workflows with it. Download JEV-9B: https://huggingface.co/autotrust/JEV-9B
This post quotes another post: x.com/huggingmodels/status/2107195374201135530