cross-layer-sharing

DepthWeave-KV

DepthWeave-KV: Token-Adaptive Cross-Layer Residual Factorization for Long-Context KV Cache Compression

跨层共享低秩基结合 token 自适应残差与输出误差反馈压缩 KV；报告 8.3× 缩减，但实验配置缺失且消融存在冲突。