ZigZagKV: Dynamic KV Cache Compression for Long-context Modeling based on Layer Uncertainty

ZigZagKV: Dynamic KV Cache Compression for Long-context Modeling based on Layer Uncertainty Meizhi Zhong author Xikai Liu author Chen Zhang author Yikun Lei author Yan Gao author Yao Hu author Kehai Chen author Min Zhang author 2025-01 text Proceedings of the 31st International Conference on Computational Linguistics Owen Rambow editor Leo Wanner editor Marianna Apidianaki editor Hend Al-Khalifa editor Barbara Di Eugenio editor Steven Schockaert editor Association for Computational Linguistics Abu Dhabi, UAE conference publication zhong-etal-2025-zigzagkv https://aclanthology.org/2025.coling-main.596/ 2025-01 8897 8907