XiaomiMiMo

/

MiMo-7B-RL-0530

like 32

Follow

Xiaomi MiMo 685

find(search for) new formula ,this is old technology , results are not best , models unable to remember (everything were teached)correctly , only many experts moe which experts are not overloaded with info work good in compare with only one mlp overloaded with everything,this old tech need external info(depend the question) into context window and will work better but current inference players are not so advanced (to auto load external info), 7B model alone (with current formula scheme)will not achieve anything more ,with more training or not nothing will change