JevMade hello@JevMade.com
← Back to guides

JevMade field notes / Teaching implementation

Open Jev

A small, untrained PyTorch implementation that makes a hypothesized shared state encoder, cache, typed heads, training step, and calibration boundary inspectable.

Original by Kye GomezGetting startedGitHub repositorySource reviewed

Before you dive in

What you’ll find in the original

  1. Encode shared state once, then attach separate Choice, Score, and Noul readout heads.
  2. Use soft target distributions to preserve disagreement, then measure calibration under distribution shift before deployment.
  3. Treat cache reuse and multi-question scaling as hypotheses until benchmarked.
Worth knowing

This is an educational Jev-inspired architecture with a hash tokenizer and random weights, not a usable model or a reproduction of TypeSafe's undisclosed architecture. Training and benchmark work remain TODOs.