AI RESEARCH

QAM-W: Joint 2D Codebook Quantization for LLM Weights via Hadamard Rotation and Activation-Aware Scaling

arXiv CS.LG

Scalar post-