Developer's own thread: https://www.reddit.com/r/LocalLLaMA/s/adp1cGZZe9 Code: https://github.com/Inovello/llama.cpp/tree/flashnext-e06 My hardware: 2x RTX 3090, Intel Ultra 7 270k Plus, 192 GB DDR5@5600 MHz Token generation speed went from 20 t/s to 49 t/s. Prompt processing speed is 140 t/s. Promp
