Post
Michael Spencer @AISupremacyNews ยท Feb 21, 2023
  • From Twitter

Xavi says: This paper is interesting in many ways. By using multimodal fine-tuning of small(er) (i.e. 1B parameter) models like T5 the authors show that they can beat GPT3.5 in visual QA tasks, surpassing visual performance in several tasks.

Recommended by 1 person
2 mentions
Share on Twitter Repost
Replies
Reply
No replies yet