Open Models Almanac
Methodsinference & servinginference quantization

single mention · inference & serving

NVFP4 quantization of MoE expert linears

MoE expert linears are converted to NVFP4 while shared experts, attention, embeddings, and early dense layers remain BF16.

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 NVFP4 re-quantization.

Filed under inference & serving › inference quantization