How is quantization of activations handled in pytorch after QAT?

quantization
16

see my answer A few questions about QConfig in quantization - #2 by HDCharles