Open Models Almanac
Methodsinference & servinginference quantization

single mention · inference & serving

Mixed-precision key/value cache quantization

A KV-cache configuration assigning different FP8 and NVFP4 precisions to keys and values on a per-layer basis.

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.

Filed under inference & serving › inference quantization