jevos-v3

Yes/no decisions on a laptop CPU in 25–110 ms: jev, one binary for Windows, Linux and macOS, now with jevos-v3 at 8-bit weights. 80.8% on 999 hand-written yes/no questions (jevos-v2 80.3%), 86.5% on 821 new ones written by another model family (jevos-v2 83.1%), and better scores: 56.5% on 2,350 held-out score questions (54%) and 61% on everyday ratings (55%). Same size and speed as jevos-v2.

jevos-v3 is the weight average of three new training runs (same 17-layer model, a training mix that adds natural questions, choice and score questions in the form jev asks them, and more arithmetic) and jevos-v2. Measured through jev itself (OpenVINO INT8).

Download the archive for your system and the model, and unpack the model into the jev folder:

tar -xzf jev-linux-x64.tar.gz                 # Windows: unzip jev-windows-x64.zip
cd jev
unzip ../jevos-v3-openvino-int8.zip           # creates model/
./jev serve                                   # Windows: jev.exe serve

File	What it is
jev-windows-x64.zip, jev-linux-x64.tar.gz (glibc 2.35+), jev-macos-arm64.tar.gz	the jev binary with OpenVINO's libraries and the licenses
jevos-v3-openvino-int8.zip	the model jev runs, as model/
jevos-v3-q4_k_m.gguf, jevos-v3-q8_0.gguf	the same model as GGUF, for llama.cpp and the tools built on it
jevos-dino.mp4	the recording of jevos playing a Dino-style game
SHA256SUMS.txt	the SHA-256 of every file here

The API, options and numbers are in the README.