Entity

FP8

8-bit floating-point precision, a numerical format used for efficient training and inference of large neural networks.