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

ArXiv CS.CL2026-05-27NEWSen作者: Preetam Sharma, Kacper Dobek

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