Installation breaks pytorch native mixed precision

[maxmax1992]

I'm running DeepLabV3 on my GTX 1650 mobile graphis card, I've had 0.20 s per forward pass before installation, however after installing apex with this command: git clone https://github.com/NVIDIA/apex cd apex pip install -v --disable-pip-version-check --no-cache-dir --global-option="--cpp_ext" --global-option="--cuda_ext" ./ Feedforward time suddenly becomes 0.70! without even using apex in the first place (using torch native mixed precision)? I think there's some bug in the code and It would be helpful if you show me how to completely remove apex from my machine. Thank you.

[eqy]

Thanks for bringing this up; would it be possible to provide a repro snippet for this issue?

In the meantime, does pip uninstall apex work for you?