Probe-and-Fetch: Dynamic KV Cache Pruning for Accelerated Long-Context Inference in Web-Scale AI Search.

@inproceedings{li-2026-probeandfetch, author = {Yuchen Li and Rui Kong and Xinran Chen and Chengzhe Zhang and Jiamin Chen and Cheng Deng and Xinyu Ma and Haojie Zhang and Tianhao Peng and Hengyi Cai and Shuaiqiang Wang and Jiashu Zhao and Yongqi Zhang and Haoyi Xiong and Jimmy Xiangji Huang and Lei Chen and Jun Wang and Dawei Yin}, editor = {Hakim Hacid and Yoelle Maarek and Francesco Bonchi and Ido Guy and Emine Yilmaz}, title = {{Probe-and-Fetch: Dynamic KV Cache Pruning for Accelerated Long-Context Inference in Web-Scale AI Search}}, booktitle = {{Proceedings of the ACM Web Conference 2026, WWW 2026, Dubai, United Arab Emirates, originally scheduled for April 13-17, 2026, rescheduled for June 29 - July 3, 2026}}, pages = {8127--8137}, publisher = {ACM}, year = {2026}, url = {https://doi.org/10.1145/3774904.3792794}, doi = {https://doi.org/10.1145/3774904.3792794} }