Neural Search Talks โ€” Zeta Alpha cover image

ColBERT + ColBERTv2: late interaction at a reasonable inference cost

Neural Search Talks โ€” Zeta Alpha

00:00

The Importance of Distillation in Model Performance

The Colbert outperforms the others. If we look at Rocket QA2, so this is a shared baseline, there's a pretty big improvement for many of the tasks. The distillation from a cross encoder seems to be very effective. Sometimes people do this and then use it to create hard negatives. They were often better than the non-truth. So I guess we see a similar trend with the Colbert outperforming the others.

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