Neural Search Talks β€” Zeta Alpha cover image

ColBERT + ColBERTv2: late interaction at a reasonable inference cost

Neural Search Talks β€” Zeta Alpha

00:00

The Painful Cost of Storage Efficiency in Academic Research

Marco: I have to say that I'm kind of surprised that the, that's such a, um, such a focus of this paper. Because in like, it doesn't affect the benchmarks course and, you know, does the community really cares about these sort of efficiency trade or not generally? Marco: It started so painful that it would have been really hard to experiment. If you were to save the 768 dimensional vectors from Colbert v one, I believe I calculated a few terabytes of disk space just to save that collection. So for Colbert v one,. it's something like 200 gigabytes from S Marco.

Transcript
Play full episode

The AI-powered Podcast Player

Save insights by tapping your headphones, chat with episodes, discover the best highlights - and more!
App store bannerPlay store banner
Get the app