single mention · inference & serving
BF16 expert activations
An ablation that converts expert activations to BF16 to assess 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 Convert expert activations to BF16, Expert activations → BF16.
Filed under inference & serving › inference quantization