1
WUSH: Near-Optimal Adaptive Transforms for LLM Quantization
提出接近最优自适应变换方法，大幅提升LLM量化精度，模型压缩新突破。
arXiv:2512.00956v3 Announce Type: replace Abstract: Quantizing LLM weights and activations is a standard approach for efficient deployment, but a few …