Dataset : llama tokenizer 4000 distribution : 85% 15% only for attributes number of epochs : 4 Progressive : no
Base model