Abstract
Long-output reasoning shifts the KV-cache bottleneck from the fixed prompt to the generated trace. Existing reasoning-cache eviction methods largely treat cached entries as routing objects, estimating whether an old key will still be read, will recur, or can be replaced. This routing-only view overlooks two effects: low-attention entries can carry large value payloads whose removal changes future predictions, and newly generated states can appear stale before later queries have had a chance to read them. We introduce AvoKV-E, a training-free eviction policy that first delays eligibility for recent states and then ranks eligible entries using candidate-normalized read pressure, key redundancy, and value-payload potential. According to empirical evaluation across different models and datasets, AvoKV-E matches or exceeds redundancy-aware, recurrence-based, and thought-adaptive eviction baselines at matched active-KV budgets, with its largest gains in the tightest-cache regime. Component and counterfactual analyses further connect these gains to delayed observation, payload-aware scoring, redundancy, and scale-robust normalization. Together, the results show that long-reasoning KV eviction should preserve not only keys that are likely to be read, but also the value payloads that sustain the reasoning trajectory.
Keywords
Publication details
- Journal
- Not available
- Open access
- Green open access
Cite this article
APA 7
Yu, H., Zhu, W., Chen, X., Wang, Z., Sang, H., Shi, H., Zhou, M., Dong, X., Huang, M., Cai, R., Wang, H., & Geramifard, A. (2026). AvoKV-E: Payload-Aware KV Cache Eviction for Long Reasoning. https://omanscience.com/en/articles/avokv-e-payload-aware-kv-cache-eviction-for-long-reasoning
MLA 9
Yu, Han, et al. "AvoKV-E: Payload-Aware KV Cache Eviction for Long Reasoning." https://omanscience.com/en/articles/avokv-e-payload-aware-kv-cache-eviction-for-long-reasoning.
Chicago (author–date)
Yu, Han, Wenhui Zhu, Xiwen Chen, Zhipeng Wang, Hejian Sang, Han Shi, Menglin Zhou, Xuanzhao Dong, Minzhou Huang, Rui Cai, Hao Wang, and Alborz Geramifard. 2026. "AvoKV-E: Payload-Aware KV Cache Eviction for Long Reasoning." https://omanscience.com/en/articles/avokv-e-payload-aware-kv-cache-eviction-for-long-reasoning.
Harvard
Yu, H., Zhu, W., Chen, X., Wang, Z., Sang, H., Shi, H., Zhou, M., Dong, X., Huang, M., Cai, R., Wang, H. and Geramifard, A. (2026) 'AvoKV-E: Payload-Aware KV Cache Eviction for Long Reasoning', Available at: https://omanscience.com/en/articles/avokv-e-payload-aware-kv-cache-eviction-for-long-reasoning.
Vancouver
Yu H, Zhu W, Chen X, Wang Z, Sang H, Shi H, et al. AvoKV-E: Payload-Aware KV Cache Eviction for Long Reasoning. https://omanscience.com/en/articles/avokv-e-payload-aware-kv-cache-eviction-for-long-reasoning
IEEE
H. Yu, W. Zhu, X. Chen, Z. Wang, H. Sang, H. Shi, M. Zhou, X. Dong, M. Huang, R. Cai, H. Wang, and A. Geramifard, "AvoKV-E: Payload-Aware KV Cache Eviction for Long Reasoning," https://omanscience.com/en/articles/avokv-e-payload-aware-kv-cache-eviction-for-long-reasoning.