The collection of Efficiency-Prioritized Token-mixer(EPT), a model architecture aiming for maximization of LLM inference efficiency.