0.0.15
remove 2-level, simplify code, thank @arankomat for consultation. lik… …ely nobody is using 2-level moe. openai likely used 1-level with 16 experts
remove 2-level, simplify code, thank @arankomat for consultation. lik… …ely nobody is using 2-level moe. openai likely used 1-level with 16 experts