jevbooks

← All projects

von

wfzyx/von

Open-source non-autoregressive System One decision model for calibrated inference.

Classification & Routing94%Runner-up: Verification & GuardrailsQuestion health measured

What it is

Von is a 395M-parameter non-autoregressive model that answers typed questions over input state with calibrated probabilities in sub-25ms, for developers needing local, deterministic decision-making without generative text.

How it uses Jev

Von is a drop-in alternative to TypeSafe Jev, not a user of Jev. It answers Choice, Noul, and Score questions over state via von.decide, returning calibrated probabilities and confidence. The README does not describe using Jev itself.

Primitives:choicescorenoul

Technique worth stealing

RLCD post-training with joint Cross-Entropy and Brier loss plus temperature scaling for calibrated probabilities.

Try it

pip install von-sdk, then call von.decide with state, choices, and instructions.

View on GitHub

judged by Jevjev-1.13.0

Evidence

Each line is one question put to Jev about the README. ≥ 0.60 reads as yes, ≤ 0.40 as no; in between Jev is not making a call.

  • Jev-centricunclear0.58
  • Shows a System One patternno0.19
  • Handles uncertaintyno0.06
  • Measuredno0.38
  • Runnableyes0.75
  • Worth recommendingno0.09
  • Model replicayes0.94
  • Problem scopescore on a 0–2 scale1.56
  • About Jevyes0.95

Signals by Jev jev-1.13.0, card written by DeepSeek V4.1 Flash from the README on 21 Sept 2026.