jevIMAGE
Running Jev's decision model locally on CPU with a llama.cpp patch
A demo shows the 421M-parameter Laya model running purely on CPU with a 1GB peak memory footprint, decision accuracy close to Jev, and no API fees.
Hikari∣LocalLLM⚡
@Hikari_07_jpArchived 2026.10.04Originally posted 2026.10.03
Hikari∣LocalLLM⚡ @Hikari_07_jpOriginally posted 2026.10.03
You can run Jev locally using the https://atomic.chat llama.cpp patch! This demo is running on the CPU. Laya has 421M parameters and requires only 1GB of memory at peak usage. ・Decision-making accuracy comparable to Jev ・Faster performance ・No API fees.
This post quotes another post: x.com/atomic_chat_hq/status/2106145162762109160
x.com/hikari_07_jp/status/2106320874718835091Archived 2026.10.04 10:05