Should We Still Pretrain Encoders with Masked Language Modeling? Paper • 2507.00994 • Published 21 days ago • 74