GitHub · by mode-io

vllm-jev

Native vLLM serving for Jev-style decision models.

mode-io's vllm-jev serves Jev-style decision models natively on vLLM, on Linux and Apple Silicon. In the repository's live run, one Valen checkpoint answers 24 structured questions about a video from eight sampled frames. Valen's PyTorch reference path took 5.31 s and vllm-jev took 1.22 s, with all 24 selected answers in agreement.

mode-io/vllm-jev on GitHub

Stars: 337
Forks: 10
Language: Python
Last push: Oct 5, 2026

Repository created Sep 24, 2026. Counts read from the GitHub API on Oct 8, 2026, and they move daily — quote them with the date.

Open the source

CUA-S1-FORMS

Parameters: 706k
Form fill: one 50 ms pass

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…

GitHubOpen source

jevlike

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…