2 ms·A single book for pretraining boosts model performance by 'less than 0.06%.'3 points by bilsbie 2y agodeleted 2y ago[deleted]