@rohanpaul_ai: atomic[.]chat just made Gemma 4 26B faster inside LLaMA.cpp. making token generation about 40% fast...