Mlm

• Masked Language Model

  Predicts randomly masked tokens from context; primary pre-training objective for bidirectional encoders like BERT.