Hacker News new | ask | show | jobs
by jakswa 1 day ago
what quantization? FP4?
1 comments
NitpickLawyer 1 day ago
The model is native mxfp4 w/ mxfp8 activations via QAT.
link
trollbridge 20 hours ago
I have some morbid curiousity about what K3 would be like in an FP16 or FP32 model.
link