jevbooks

← All projects

jev-forge

zwliJay/jev-forge · Homepage

Open stack for training and serving Jev-style decision models.

Agent Decisions76%Runner-up: Scoring & RankingQuestion health measured

What it is

JevForge is an end-to-end toolkit for synthesizing decision data, training calibrated candidate scorers, evaluating them, and serving Jev-compatible inference for interactive web decisions. It is for researchers and developers building structured decision models.

How it uses Jev

JevForge serves a Jev-compatible API that accepts a state and typed questions (choice, noul, score). It expands each question into K candidate rows, scores each with a two-layer GELU scorer, and normalizes the K logits to return a probability distribution over candidates.

Primitives:choicenoulscore

Technique worth stealing

Pool the final valid token, apply a shared two-layer GELU scorer to each candidate, and normalize only the K logits per question.

Try it

bash scripts/run_mac_demo.sh

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-centricyes0.90
  • Shows a System One patternunclear0.50
  • Handles uncertaintyno0.06
  • Measuredno0.23
  • Runnableyes0.72
  • Worth recommendingunclear0.59
  • Model replicayes0.95
  • Problem scopescore on a 0–2 scale1.01
  • About Jevyes0.97

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