view reply Hi, nice work and interesting result :).Did you compare these with a training on x2 and x4 epoch on a baseline model to benchmark the deviation of a "standard" method?