Follow
Mask More and Mask Later: Efficient Pre-training of Masked Language Models by Disentangling the [MASK] Token. (arXiv:2211.04898v2 [cs.CL] UPDATED)
Mask More and Mask Later: Efficient Pre-training of Masked Language Models by Disentangling the [MASK] Token. (arXiv:2211.04898v2 [cs.CL] UPDATED)