single source · inference & serving
Promote expert weights from NVFP4 to FP8
An ablation that raises routed and shared expert weight precision from NVFP4 to FP8 to examine quantization-related behavior.
The sources on this page only mention it in passing. The line above is the extraction's one-line gloss. See inference quantization for the methods filed beside it.
Also called Routed/shared experts NVFP4→ FP8.
Filed under inference & serving › inference quantization