About BatchNorm QAT 's problems

microsoft / nni

An open source AutoML toolkit for automate machine learning lifecycle, including feature engineering, neural architecture search, model compression and hyper-parameter tuning.

https://nni.readthedocs.io

MIT License

14.06k stars 1.82k forks source link

About BatchNorm QAT 's problems #5704

Open typhoonlee opened 1 year ago

typhoonlee commented 1 year ago

Describe the bug: when i use QAT_Quantizer() to qat my model，i got a wrap_model.pt，then i test with this model，encounter this bug：

Environment: linux

NNI version: version = '999.dev0'
Training service (local|remote|pai|aml|etc): remote
Python version: 3.7.10
PyTorch version: 1.12.0 + cu113
Cpu or cuda version: GPU version GeForce RTX 3090

Reproduce the problem

Code|Example:

and my model arch is :

How to reproduce: