open-jev (daseinlabs)
One-pass option scoring with a local Gemma 3 4B on Apple silicon via MLX, inspired by jevlike, with a Doom demo.
Open Jev implementation with custom finetuning
How it uses Jev
demo/doom/ runs ViZDoom headless, describes each frame in a line of text, and lets the server rank the action menu with one /score call (or one System One choice question). Start make serve , then make doom . Details and keys in demo/doom/README.md .
Highlights
- openjev features DATA --out F.npz stops Gemma before the LM head, keeps every context token's
- openjev train = AdamW 5e-4 (2e-3 diverges on Gemma features, whose norms are 115), weight decay 1e-4, grad clip 1.0, 8 epochs, batch 64, best validation
- openjev eval-head reports what jevlike-eval reports: top-1, top-3, 10-bin expected calibration
Getting started
For use inside a Python application or notebook, install this checkout with python -m pip install . (Python 3.12+). See the Python package guide for decision inference, feature extraction, training, and wheel installation.
Source: community catalog. Details and metrics reflect the published community snapshot and are not a performance endorsement. Documentation details extracted from GitHub · the README · the project website.