Language Model: Roberta, T5

Ko Hyejung·2021년 12월 5일
0

2021 SKT AI

목록 보기
12/12
post-custom-banner

Roberta

Additional optimization of Bert

  • large batch size
  • no more 'next sentence prediction' loss
  • 10x more data (160GB w/common crawl)
  • more compute: 1,000 v100 GPU

T5 (Text-to-Text Transfer Transformer)

post-custom-banner

0개의 댓글