Neural Search Talks — Zeta Alpha cover image

ColBERT + ColBERTv2: late interaction at a reasonable inference cost

Neural Search Talks — Zeta Alpha

00:00

Hertz Performance Drop

It's something like learning the right version of a term and then having a small delta to tweak that a little bit. So this is like photo that goes with artist and then we use that delta vector to slightly tweak it, which I guess can only make that each dimension a bit bigger or smaller. It can't change anything too much because you're saving one or two bits. Yeah. Well, I'm going to say, would you expect to have a pretty big implementation kind of like complexity? All right. Sounds reasonable.

Transcript
Play full episode

Remember Everything You Learn from Podcasts

Save insights instantly, chat with episodes, and build lasting knowledge - all powered by AI.
App store bannerPlay store banner