specific method · filed under inference & serving
MXFP4 weight quantization
Stores weights in microscaled 4-bit floating point with per-block scaling factors.
Also called Microscaling FP4 (MXFP4) weights, MXFP4 weights, native MXFP4 quantization.
- sources
- 3
- models
- 2
- labs adopt it
- 2
- strongest
- used
How sources treat it
One count per evidence span, weakest treatment to strongest.
used 3
Documented in
Further reading
Picked by hand, not extracted: where to read more, not evidence for anything on this page.
- Microscaling Data Formats for Deep Learning paper arxiv.orgintroduces the MX format family including MXFP4
- OCP Microscaling Formats (MX) Specification v1.0 docs www.opencompute.orgthe formal MXFP4 spec (E2M1, block-32, E8M0 scale)
Evidence
3 spans quoted from the sources, strongest treatment first.
using MXFP4 weights
usedtraining objectivein Kimi K3Moonshot AI
MXFP4 weights (Microscaling FP4): Each weight is stored in 4-bit floating point with per-block scaling factors.
usedmodel architecturein Kimi K3Moonshot AI
is optimized to run on a single H100 GPU with native MXFP4 quantization
usedinference servingin gpt-oss-120bOpenAI
Filed alongside
Other methods under inference & serving :: inference quantization.
FP8 KV-cache quantizationFP8FP4 KV-cache quantizationNVFP4 quantizationFour-Over-SixPost-training quantizationQuantizationBlock-scaled INT8 quantization with stochastic roundingFP8 E4M3 quantizationGGUFMax-based scalingMSE-based scalingMXFP8 activation quantizationNVFP4 KV-cache quantizationNVFP4 ModelOpt re-quantizationSSM cache quantizationAWQ INT4 weight quantization (W4A16)BF16 inferenceBlock-wise E4M3 FP8 weight quantizationChannel-wise quantizationDynamic activation scalingEmbedding and KV-cache quantizationEmpirical bits-per-element budget selectionFine-grained FP8 quantization