Remix.run Logo
Remix clone Hacker News
new | show | ask | jobsGithub
▲ mchiang 7 days ago

OpenAI has only provided MXFP4 weights. These are the same weights used by other cloud providers.

▲irthomasthomas 7 days ago | parent [-]

Oh, I didn't know that. Weird!

▲reissbaker 7 days ago | parent [-]

It was natively trained in FP4. Probably both to reduce VRAM usage at inference time (fits on a single H100), and to allow better utilization of B200s (which are especially fast for FP4).

▲irthomasthomas 7 days ago | parent [-]

Interesting, thanks. I didn't know you could even train at FP4 on H100s

▲reissbaker 5 days ago | parent [-]

It's impressive they got it to work — the lowest I'd heard of this far was native FP8 training.