YFarmX

INT8 quantisation

AI

INT8 quantisation: Storing model weights, and sometimes activations, as 8-bit integers instead of 16- or 32-bit numbers, cutting memory use with little loss of accuracy.

Related terms

Browse the full glossary →