What it does
The project trains and evaluates Gemma 3 270M and Gemma 4 E2B models on Modal.
Benchmarks & research
system-one-open recreates Jev-style calibrated decisions with small open Gemma models.
Screenshot unavailable. Open the experiment ↗
The project trains and evaluates Gemma 3 270M and Gemma 4 E2B models on Modal.
For a smart-home app, start with the example that turns a written request into a device choice and an action. A developer can adapt it to ask about unclear requests and whether confirmation is needed. This project runs on Modal, a service that provides computing over the internet. Treat the AI's choices as suggestions to check, not proof that an action is safe.