[caffe2] Post train quantization

[zetyquickly (Emil Bogomolov)]

Hello there,

Since pytorch==1.3 has released there is an ability to perform quantization of the model during training time. And as I understand quantization works with QNNPACK.
Before that we had a way to run caffe2 models using QNNPACK too.

The question is following is there a way to quantize caffe2 model (already trained) today?

[jerryzh168 (Jerry Zhang)]

@zetyquickly

The question is following is there a way to quantize caffe2 model (already trained) today?

We have internal script to quantize these models but it’s not open sourced I think. Our recommendation is to switch to PyTorch