Similar Items: A Scalable Recipe on SuperMUC-NG Phase 2: Efficient Large-Scale Training of Language Models