⬢github Python · 609 ★ +2 since we first saw it · pushed 2 d ago · Apache-2.0
Liuziyu77/Valen
Train a Jev-like multimodal model by yourself. System One Model, now with vision.
Valen is an open-source multimodal 'System One' decision model inspired by Jev. It takes text, images, and video plus a set of candidate options, and outputs probabilities over those candidates using a small Qwen backbone with a decision head — no text generation. The repo provides the model code, training datasets, SFT and experimental RLCD training, evaluation, and a preview model with demo.
Why now: It's trending as a newly released, most-starred repo, with a demo and a striking benchmark showing a 0.8B/2B decision model beating a 27B generation model on a game puzzle at far lower latency.
Who it is for: ML researchers and developers who want to train or use a small multimodal model that picks among candidate options instead of generating free-form text.
Stars over our 9 snapshots: 607 to 609, since 2 h ago.
Where people talked about it
- ⬢github new repos, most starred 5 min ago
API: https://socialmediatrends-api.osmike.com/v1/repos/Liuziyu77/Valen