implementation detail · filed under optimization
MXFP4 Weights with MXFP8 Activations
A mixed-precision configuration using MXFP4 for weights and MXFP8 for activations, described alongside QAT but not establishing a distinct training algorithm.
Also called MXFP4/MXFP8 quantization-aware training.
- sources
- 2
- model
- 1
- lab adopt it
- 1
- strongest
- used
How sources treat it
One count per evidence span, weakest treatment to strongest.
Documented in
Further reading
Picked by hand, not extracted: where to read more, not evidence for anything on this page.
- Kimi K3: Open Frontier Intelligence paper arxiv.orgMXFP4 weights / MXFP8 activations, QAT from SFT onward
Evidence
3 spans quoted from the sources, strongest treatment first.
MXFP4 weights / MXFP8 activations (quantization-aware training)
Kimi K3 applies quantization-aware training from the SFT stage onward, using MXFP4 weights with MXFP8 activations for broad hardware compatibility.
Kimi K3 applies quantization-aware training from the SFT stage onward, using MXFP4 weights with MXFP8 activations for broad hardware compatibility.
Filed alongside
Other methods under optimization :: quantization-aware training.