What is the recommended way to save the PyTorch model with quantizer information?

[RyougiKukoc]

As I understand, the state dict of the model does not contain any information about the quantizer (such as scale and zeropoint). So what is the recommended way to save model in QAT? Just export to onnx every time?

[nickfraser]

the state dict of the model does not contain any information about the quantizer (such as scale and zeropoint).

This is incorrect. The state_dict does contain this information when this information is not generated "on-the-fly", in which case it is not necessary. To be clear, you can do something like this:

torch.save(quant_model.state_dict(), 'checkpoint.pt')
quant_model.load_state_dict(torch.load('checkpoint.pt'))

If this is not what you are trying to do, can you explain it in more detail?