Re:Open models by OpenAI
by rushingcreek on Tuesday August 05, 2025 @03:11PM
(#44802813)
The native FP4 is one of the most interesting architectural aspects here IMO, as going below FP8 is known to come with accuracy tradeoffs. I'm curious how they navigated this and how the FP8 weights (if they exist) were to perform.
They said it was native FP4, suggesting that they actually trained it like that; it's not post-training quantisation.