power-efficient

PrismML's Bonsai 27B model runs on the Jetson Orin Nano 8GB with 4.31 tokens/s and 27 t/s prompt processing, using 6.2GB RAM and about 25W power. It indicates surprisingly usable edge AI performance.