how are brevitas layers trained?

[rnd810]

Dear all,

I can't find any mention about how brevitas layers are trained. Does model use Straight Through Estimator(STE) when it encounters quantization layer in its backward path?

[seniorgoogles]

Dear all,

I am also interested in this topic. How does the QAT work here? Is quantization applied in forward and backward pass like in Pytorch or only in forward path?

Kind regards,
Michael

[Giuseppe5]

Yes, we use STE during QAT