ParisKV: Fast and Drift-Robust KV-Cache Retrieval for Long-Context LLMs

Yanlin Qi, Xinhang Chen, Huiqiang Jiang, Qitong Wang, Botao Peng, Themis Palpanas
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:100229-100248, 2026.

Abstract

KV-cache retrieval is essential for long-context LLM inference, yet existing methods struggle with distribution drift and high latency at scale. We introduce ParisKV, a drift-robust, GPU-native KV-cache retrieval framework based on collision-based candidate selection, followed by a quantized inner-product reranking estimator. For million-token contexts, ParisKV supports CPU-offloaded KV caches via Unified Virtual Addressing (UVA), enabling on-demand top-$k$ fetching with minimal overhead. ParisKV matches or outperforms full attention quality on both long-input and long-generation benchmarks. It achieves state-of-the-art long-context decoding efficiency: it matches or exceeds full-attention speed even at batch size 1 for long contexts, delivers up to 2.8$\times$ higher throughput within full attention’s runnable range, and scales to million-token contexts where full attention runs out of memory. At million-token scale, ParisKV reduces decode latency by 17$\times$ and 44$\times$ compared to MagicPIG and PQCache, respectively—two state-of-the-art KV-cache top-$k$ retrieval baselines; code is available at https://github.com/amy-77/ParisKV/tree/main.

Cite this Paper

BibTeX
@InProceedings{pmlr-v306-qi26b, title = {{P}aris{KV}: Fast and Drift-Robust {KV}-Cache Retrieval for Long-Context {LLM}s}, author = {Qi, Yanlin and Chen, Xinhang and Jiang, Huiqiang and Wang, Qitong and Peng, Botao and Palpanas, Themis}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {100229--100248}, year = {2026}, editor = {Zhang, Tong and Dudik, Miroslav and Jaggi, Martin and Agarwal, Alekh and Li, Sharon and Schuurmans, Dale and Zhu, Jerry and Berkenkamp, Felix and Dong, Hanze and Bietti, Alberto}, volume = {306}, series = {Proceedings of Machine Learning Research}, month = {06--11 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v306/main/assets/qi26b/qi26b.pdf}, url = {https://proceedings.mlr.press/v306/qi26b.html}, abstract = {KV-cache retrieval is essential for long-context LLM inference, yet existing methods struggle with distribution drift and high latency at scale. We introduce ParisKV, a drift-robust, GPU-native KV-cache retrieval framework based on collision-based candidate selection, followed by a quantized inner-product reranking estimator. For million-token contexts, ParisKV supports CPU-offloaded KV caches via Unified Virtual Addressing (UVA), enabling on-demand top-$k$ fetching with minimal overhead. ParisKV matches or outperforms full attention quality on both long-input and long-generation benchmarks. It achieves state-of-the-art long-context decoding efficiency: it matches or exceeds full-attention speed even at batch size 1 for long contexts, delivers up to 2.8$\times$ higher throughput within full attention’s runnable range, and scales to million-token contexts where full attention runs out of memory. At million-token scale, ParisKV reduces decode latency by 17$\times$ and 44$\times$ compared to MagicPIG and PQCache, respectively—two state-of-the-art KV-cache top-$k$ retrieval baselines; code is available at https://github.com/amy-77/ParisKV/tree/main.} }
Endnote
%0 Conference Paper %T ParisKV: Fast and Drift-Robust KV-Cache Retrieval for Long-Context LLMs %A Yanlin Qi %A Xinhang Chen %A Huiqiang Jiang %A Qitong Wang %A Botao Peng %A Themis Palpanas %B Proceedings of the 43rd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2026 %E Tong Zhang %E Miroslav Dudik %E Martin Jaggi %E Alekh Agarwal %E Sharon Li %E Dale Schuurmans %E Jerry Zhu %E Felix Berkenkamp %E Hanze Dong %E Alberto Bietti %F pmlr-v306-qi26b %I PMLR %P 100229--100248 %U https://proceedings.mlr.press/v306/qi26b.html %V 306 %X KV-cache retrieval is essential for long-context LLM inference, yet existing methods struggle with distribution drift and high latency at scale. We introduce ParisKV, a drift-robust, GPU-native KV-cache retrieval framework based on collision-based candidate selection, followed by a quantized inner-product reranking estimator. For million-token contexts, ParisKV supports CPU-offloaded KV caches via Unified Virtual Addressing (UVA), enabling on-demand top-$k$ fetching with minimal overhead. ParisKV matches or outperforms full attention quality on both long-input and long-generation benchmarks. It achieves state-of-the-art long-context decoding efficiency: it matches or exceeds full-attention speed even at batch size 1 for long contexts, delivers up to 2.8$\times$ higher throughput within full attention’s runnable range, and scales to million-token contexts where full attention runs out of memory. At million-token scale, ParisKV reduces decode latency by 17$\times$ and 44$\times$ compared to MagicPIG and PQCache, respectively—two state-of-the-art KV-cache top-$k$ retrieval baselines; code is available at https://github.com/amy-77/ParisKV/tree/main.
APA
Qi, Y., Chen, X., Jiang, H., Wang, Q., Peng, B. & Palpanas, T.. (2026). ParisKV: Fast and Drift-Robust KV-Cache Retrieval for Long-Context LLMs. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:100229-100248 Available from https://proceedings.mlr.press/v306/qi26b.html.

Related Material