Vector compression cuts memory use 80,000x with no accuracy loss

Original: tpuf quantizes vectors to improve perf (RaBitQ)

Source: x.com ↗