LLM Paper Club (RecSys with Generative Retrieval (RQ-VAE))

​Eugene Yan senpai is back!

This week, we'll go through a paper at the intersection of llms and recsys. I think the technique it proposes—having an RQ-VAE convert embeddings to tokens—is applicable to many use cases, and I know that several recsys teams are experimenting with it. Please read the paper. If you have time, try to grok section 3.1 with your fav AI teacher.

​

​We'll go through (i) the paper, (ii) code implementation, (iii) rq-vae training runs, and (iv) sasrec vs. sasrec + semantic IDs, depending on time. https://arxiv.org/abs/2305.05065

​---

​we need YOU to volunteer to do rapid-fire recaps and explanations of our remaining papers on the board: https://app.sli.do/event/bNV6mo3BFGhe8Bqzb1tonb/live/questions

​please sign up in #llm-paper-club in https://www.latent.space/p/community discord

​recordings at /LatentSpaceTV