Hello, AIMET team!
Working on Resnet101’s QAT. I’m attempting to replicate the AIMET github code for QAT (Examples folder).
PyTorch Resnet101:
The PTQ works correctly when using the same script as in the examples, but the QAT accuracy reduced to 0.096% after 1 epoch (top1). And after one epoch, it gets terminated without any error or stack trace in the middle of the second epoch.
TF-Keras Resnet 101:
Similar to PyTorch, the accuracy dropped to 0.008% after one epoch (top1).
Could you please suggest on how to proceed?