humanizer v2 training data: SFT, DPO and RL rows used to train v2, split by license

initial commit