jevbooks

← All projects

jevbetter

olanotolu/jevbetter

One-pass scorer over changing text options, stronger than jevlike.

What it is

jevbetter is a from-scratch Python one-pass option scorer: given a context and a variable list of text options, it returns one probability per option. It targets routing, ranking, open-label classification, and game controllers.

How it uses Jev

It is a Jev-like model, not Jev itself. It scores each option in a single forward pass using a hashed n-gram encoder, rival-aware attention among options, and a gated MLP head, then a temperature-scaled softmax yields probabilities. The README does not describe using the actual Jev model or its API.

Technique worth stealing

Rival-aware attention: options attend to each other before scoring, so near-miss alternatives sharpen the call.

Try it

pip install -e '.[dev]', then jevbetter-data, jevbetter-train, jevbetter-eval, jevbetter-predict.

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-centricyes0.70
  • Shows a System One patternno0.19
  • Handles uncertaintyno0.04
  • Measuredno0.15
  • Runnableno0.33
  • Worth recommendingunclear0.40
  • Model replicayes0.97
  • Problem scopescore on a 0–2 scale1.03
  • About Jevyes0.88

Signals by Jev jev-1.13.0, card written by DeepSeek V4.1 Flash from the README on 20 Sept 2026.