the precision a4w4 of training MobilenetV2 is nearly 0 about apot_quantization HOT 4 OPEN

Hi,

training MobilenetV2 requires you to implement (signed) asymmetric quantization for activations. Since the last layer of the inverted Residual Bottleneck block does not have ReLU function, therefore, its activation is signed.

Thanks.

from apot_quantization.

Thanks for your reply! Is it possible (how?) to use per-channel weight quantization in your APOT to boost the performance?

from apot_quantization.

Dear talenz, how about your quantization result on mbv2?

from apot_quantization.

@talenz

from apot_quantization.