How to use custom data types of Tensor

(Qing Yang) #1

I want to do 8bit quantization, but not use int8. Can I use 8bit which like 4 bits being integers and 4 bits floating point numbers?
Can anyone help me