What is the state of the art on putting text samples into a latent space?

3 points · 2 comments · view on lemmy.world

2 Comments

gmatkins@sh.itjust.works · 2 pts · 3y (1 reply)

Used to be Bert. Might still be Bert. Everybody likes Bert.

Wow. That's pretty vintage and tiny by ML standards, but it looks like it is still relevant.

There's also Phrase-BERT if you want an embedding of the whole phrase. What are some models for getting the things back out?