Adaptive residual scaling improves low-bit model compression

Original: MARR: Module-Adaptive Residual Reconstruction for Low-Bit Post-Training Quantization

Source: arxiv.org ↗