Language Model (1) 썸네일형 리스트형 [논문이해] Should You Mask 15% in Masked Language Modeling? 논문명: Should You Mask 15% in Masked Language Modeling? 논문링크: https://arxiv.org/abs/2202.08005 Should You Mask 15% in Masked Language Modeling? Masked language models (MLMs) conventionally mask 15% of tokens due to the belief that more masking would leave insufficient context to learn good representations; this masking rate has been widely used, regardless of model sizes or masking strategies. In .. 이전 1 다음
