Re:GPT-4 API General Availability
by renewiltord on Thursday July 06, 2023 @04:46PM
(#36622731)
Probably compute-bound for inference which they've probably built in an arch-specific way, right? This sort of thing happens. You can't use AVX-512 in Alibaba Cloud cn-hongkong, for instance, because there's no processor available there that can reliably do that (no Genoa CPUs there). I imagine OpenAI has a similar constraint here.
[ Parent
| on HN ]
There are no comments on this comment yet.