What is quantization-aware training (QAT) in PyTorch?

QAT simulates quantization during training to preserve accuracy in INT8 models.

QAT inserts fake quantization modules during training to mimic low-precision inference behavior, allowing model to adapt weights accordingly before final conversion.