Comment by irthomasthomas

8 days ago

Oh, I didn't know that. Weird!

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).