Laya 本地小模型跑毫秒级决策，比 Jev 快 50 倍

moctx @moc_tx89原推发布于 2026.09.21

在 X 上打开原推

Laya, this speed and size are honestly a bit ridiculous After watching @mizorewww 's demo 50 times faster than Jev running on your device laya mlx! Laya only has about 0.4B parameters, memory usage during inference is just around 1GB making millisecond level decisions right on your own device For production scenarios with high frequency decisions This kind of low latency and low resource usage is way too appealing No need to wait for cloud round trips every time, reducing reliance on networks and remote services Real time interactions, workflow routing definitely worth integrating into business to try out

这条是引用着另一条写的：x.com/moc_tx89/status/2101575650821955840