implementation detail · filed under optimization
FP4+FP8 mixed precision
MoE expert parameters use FP4 while most other parameters use FP8.
Also called FP4 + FP8 Mixed.
- sources
- 2
- models
- 2
- lab adopt it
- 1
- strongest
- used
How sources treat it
One count per evidence span, weakest treatment to strongest.
used 2
Documented in
Evidence
2 spans quoted from the sources, strongest treatment first.
MoE expert parameters use FP4 precision; most other parameters use FP8.
usedinference servingin DeepSeek-V4-FlashDeepSeek
FP4 + FP8 Mixed: MoE expert parameters use FP4 precision; most other parameters use FP8.
usedinference servingin DeepSeek-V4DeepSeek
Filed alongside
Other methods under optimization :: training precision.
NVFP4BF16NVFP4 pre-trainingFP8 mixed-precision trainingFP8-precision reinforcement learningMXFP8BF16 gradient reductionBF16 mixed-precision trainingBlock-wise FP8 activation quantization with offloadE2M1FP32 attention-output retentionFP32 gradient reductionFP8 storage for the residual stateHigh-precision final network layersMixed-FP8 quantizationMixed-precision trainingNVFP4 fine-grained micro-block scalingTwo-dimensional block quantization