GitHub · by Heman10x
A 151M non-autoregressive decision engine, with its numbers published.
The second version of the open Verdict engine. It is a 151 million parameter non-autoregressive decision model, and the repository reports 77.10% accuracy, a Brier score of 0.0636 and an expected calibration error of 0.0144 on their typed-decisions set, which it says beats Jev and Laya.
Open the source
Vincent Wang-Maścianica
@vinnylarouge
I reverse-engineered a jev-like architecture given its type. You can find the repo here to train your own jevlikes: github.com/vinnylarouge/j…
GitHubResearch and data
Eric Zhang
@ekzhang1
Inspired by @typesafeai , here is a Jev-compatible public API to play with It runs a comparable open model (Qwen3.6-35B-A3B), and just uses SGLang radix cache to preserve the prefill reuse / really fast parallel systemone generation - 64 tasks in <1s. github.com/ekzhang/openje…
GitHubResearch and data
A small open decision model: state and typed questions in, probabilities out.
Kshetrajna Raghavan
57