3 ms·

what quantization? FP4?

• NitpickLawyer 2mo ago
  The model is native mxfp4 w/ mxfp8 activations via QAT.
  • trollbridge 2mo ago
    I have some morbid curiousity about what K3 would be like in an FP16 or FP32 model.