AI RESEARCH
QAM-W: Joint 2D Codebook Quantization for LLM Weights via Hadamard Rotation and Activation-Aware Scaling
arXiv CS.LG
•
Scalar post-