How can i Using (Mixed Precision/ Automatic Mixed Precision) INT8(IMMA)

AI & Data Science Deep Learning (Training & Inference)
1

Hi, I want to apply MP/AMP (Mixed Precision/ Automatic Mixed Precision) on my model.

I used apex.amp library to change FP32 to FP16 on my model and succeeded in changing it.

Now, I want to change FP32(HMMA) to INT8(IMMA) But I couldn’t get any information.

Nvidia supports INT8(IMMA)library…?