Learning to Evict from Key-Value Cache
Fuente:
arXiv
Guardado en:
| Autores principales: | , , |
|---|---|
| Formato: | Preprint |
| Publicado: |
2026
|
| Materias: | |
| Acceso en línea: | |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| _version_ | 1866915790430666752 |
|---|---|
| author | Moschella, Luca Manduchi, Laura Sener, Ozan |
| author_facet | Moschella, Luca Manduchi, Laura Sener, Ozan |
| contents | The growing size of Large Language Models (LLMs) makes efficient inference challenging, primarily due to the memory demands of the autoregressive Key-Value (KV) cache. Existing eviction or compression methods reduce cost but rely on heuristics, such as recency or past attention scores, which serve only as indirect proxies for a token's future utility and introduce computational overhead. We reframe KV cache eviction as a reinforcement learning (RL) problem: learning to rank tokens by their predicted usefulness for future decoding. To this end, we introduce KV Policy (KVP), a framework of lightweight per-head RL agents trained on pre-computed generation traces using only key and value vectors. Each agent learns a specialized eviction policy guided by future utility, which evaluates the quality of the ranking across all cache budgets, requiring no modifications to the underlying LLM or additional inference. Evaluated across two different model families on the long-context benchmark RULER and the multi-turn dialogue benchmark OASST2-4k, KVP significantly outperforms baselines. Furthermore, zero-shot tests on standard downstream tasks (e.g., LongBench, BOOLQ, ARC) indicate that KVP generalizes well beyond its training distribution and to longer context lengths. These results demonstrate that learning to predict future token utility is a powerful and scalable paradigm for adaptive KV cache management. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2602_10238 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Learning to Evict from Key-Value Cache Moschella, Luca Manduchi, Laura Sener, Ozan Computation and Language Machine Learning The growing size of Large Language Models (LLMs) makes efficient inference challenging, primarily due to the memory demands of the autoregressive Key-Value (KV) cache. Existing eviction or compression methods reduce cost but rely on heuristics, such as recency or past attention scores, which serve only as indirect proxies for a token's future utility and introduce computational overhead. We reframe KV cache eviction as a reinforcement learning (RL) problem: learning to rank tokens by their predicted usefulness for future decoding. To this end, we introduce KV Policy (KVP), a framework of lightweight per-head RL agents trained on pre-computed generation traces using only key and value vectors. Each agent learns a specialized eviction policy guided by future utility, which evaluates the quality of the ranking across all cache budgets, requiring no modifications to the underlying LLM or additional inference. Evaluated across two different model families on the long-context benchmark RULER and the multi-turn dialogue benchmark OASST2-4k, KVP significantly outperforms baselines. Furthermore, zero-shot tests on standard downstream tasks (e.g., LongBench, BOOLQ, ARC) indicate that KVP generalizes well beyond its training distribution and to longer context lengths. These results demonstrate that learning to predict future token utility is a powerful and scalable paradigm for adaptive KV cache management. |
| title | Learning to Evict from Key-Value Cache |
| topic | Computation and Language Machine Learning |
| url | https://arxiv.org/abs/2602.10238 |