flâneur — a map of the web's best reading

RETRO Is Blazingly Fast | Mitchell A. Gordon

mitchgordon.me · 1,701 words · saved by 1 readers

When I first read Google’s RETRO paper, I was skeptical. Sure, RETRO models are 25x smaller than the competition, supposedly leading to HUGE savings in training and inference costs. But what about the new trillion token “retrieval database” they added to the architcture? Surely that must add back some computational costs, balancing the cosmic seesaw?

When I first read Google’s RETRO paper, I was skeptical. Sure, RETRO models are 25x smaller than the competition, supposedly leading to HUGE savings in training and inference costs. But what about the new trillion token “retrieval database” they added to the architcture? Surely that must add back some computational costs, balancing the cosmic seesaw? Apparently not. After running benchmarks for myself, at scale, I am convinced that RETRO is indeed BLAZINGLY fast. RETRO is so fast and cheap, in fact, that I cannot fathom why anyone would choose to do language modeling without retrieval. RETRO O

Explore this link on the map →

related reading