jevbetter
A stronger one-pass scorer over a variable list of text options: hashed n-gram encoder, rival-aware attention, gated head, temperature scaling, benchmarked against jevlike.
A stronger one-pass scorer over a variable list of text options. Hashed n-gram encoder, rival-aware attention, gated head, temperature scaling — with a head-to-head benchmark vs the jevlike starter design.
How it uses Jev
Independent project. Not affiliated with TypeSafe or the Jev model.
Getting started
The evaluation prints top-1/3/5 accuracy, MRR, calibration error, accuracy by menu size, throughput, and a shuffled-context control (each menu paired with the wrong context, which a useful model beats comfortably).
Source: community catalog. Details and metrics reflect the published community snapshot and are not a performance endorsement. Documentation details extracted from GitHub · the README.