DFFG: Fast Gradient Iteration for Data-free Quantization
We propose DFFG, a data-free quantization method based on fast gradient iteration, which leverages a momentum-based FGSM strategy to efficiently generate data for model quantization.
Nov 1, 2023
