Beyond Shadow Weights: Quantization-Aware Training as Quantized-Endpoint Descent

Sheng-An Xu, Hanyang Li, Jianhao Ma +1

Quantization-aware training (QAT) updates a full-precision shadow weight $\mathbf{x}$ but deploys the quantized endpoint $Q(\mathbf{x})$. Existing explanations for QAT largely view…