jevIMAGE
开源复刻 Jev 的 Laya:本地跑、零 API 费用
印度开发者做了 Apache 2.0 的 Jev 同类项目 Laya,同样是「状态进、结构化决策出」,在 typed-decisions 上 0.766 对 0.727,GPU 上更快且不花 API 钱,但在大标签集和未微调场景下落后。
Kunal
@kunal_twts2026.09.23收录原推发布于 2026.09.22



Kunal @kunal_twts原推发布于 2026.09.22
an Indian built an open-source Jev and it’s barely trending. It is called Laya same job as Jev: state in, typed decision out. Apache 2.0. runs on your machine. published benches where it leads: typed-decisions 0.766 vs 0.727 faster on GPU $0 instead of an API bill it loses on huge label sets and without fine-tune. Someone also did a comparison on a video rather than just plain sides
x.com/kunal_twts/status/2102371859270660546采集于 2026.09.23 10:17