原书出处
Natural Language Processing in Action, Second Edition
During pretraining, a percentage of input tokens are masked randomly (with a [MASK] token), and the model predicts the actual token IDs for those masked tokens. In practice, 15% of the WordPiece tokens were selected to be masked for training; however, a downside of this is that during fine-tuning, there is no [MASK] token. To work around this, the authors came up with a formula
原书定位详情
来源:a19c7933-4eda-4176-aebd-30fe7e1ad292
正文位置:extracted/Natural Language Processing in Action, Second Edition/_epub-src/OEBPS/Text/chapter-9.html
EPUB SHA-256:0e9a35e49ca446e322e5524e53f39fb6bee5b2b23d9cda10ef052aca770b0b36