Jeff: Home-Trained 0.8B Decision Models Running at ~30 ms

firelex · hn · 2026-09-29

Jeff is an open-source GitHub project offering Jev-compatible 0.8B decision models that the author claims can be trained at home with roughly 30 ms inference, targeting low-latency decision tasks on local hardware.

Original post →

More from coding & agent

coding & agent channel →