ParentFull threadThomashuet·Summary: We've use the most complexest, buzzwordiest training infrastructure to increase the performance of our base model by a whopping 0.5% (±1%).View on HN