DoReMi: Optimizing Data Mixtures Speeds Up Language Model Pretraining

Open in new window