arXiv - CSCL: "Mask More and Mask Later: Efficient Pre-training …" - CleverLibre Social

arXiv - CSCL @arxiv_cscl@social.cleverlibre.org

Mask More and Mask Later: Efficient Pre-training of Masked Language Models by Disentangling the [MASK] Token. (arXiv:2211.04898v1 [cs.CL])

http://arxiv.org/abs/2211.04898

Nov 10, 2022, 18:02 · · arxiv-cscl · · ·

Sign in to participate in the conversation