method · first seen 8 Oct 2026
cross-layer cache sharing
Also: cross-layer KV sharing, cross-layer KV cache sharing
Exploiting similarity between cached states of different transformer layers to reduce memory usage.
Mentions per week
PapersCodeDiscussion
Earliest sources
- VFold: Symmetry-Aware Cross-Layer Value Cache CompressionPapers · 8 Oct 2026 · Neha Verma, Sungwon Kim, Kenton Murray, Kevin Duh
- BellowsKV: One Latent for Many Layers, Expanded on DemandPapers · 26 Oct 2026 · Jiamu Zhang, Hanjie Chen
- DepthGraft: Structural Regularization through Hierarchical Cross-Layer KV ReconstructionPapers · 2 Nov 2026 · Xiaohan Qin, Xiangdong Zhang, Yu Wang, Huaijin Wu, Zhuo Xia, Yebin Yang, Junchi…
Latest sources
- DepthGraft: Structural Regularization through Hierarchical Cross-Layer KV ReconstructionPapers · 2 Nov 2026 · Xiaohan Qin, Xiangdong Zhang, Yu Wang, Huaijin Wu, Zhuo Xia, Yebin Yang, Junchi…
- BellowsKV: One Latent for Many Layers, Expanded on DemandPapers · 26 Oct 2026 · Jiamu Zhang, Hanjie Chen
- VFold: Symmetry-Aware Cross-Layer Value Cache CompressionPapers · 8 Oct 2026 · Neha Verma, Sungwon Kim, Kenton Murray, Kevin Duh