KV Packet: Recomputation-Free Context-Independent KV Caching for LLMs
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2026
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866914481846616064 |
|---|---|
| author | Chen, Chuangtao Zhang, Grace Li Yin, Xunzhao Zhuo, Cheng Li, Bing Schlichtmann, Ulf |
| author_facet | Chen, Chuangtao Zhang, Grace Li Yin, Xunzhao Zhuo, Cheng Li, Bing Schlichtmann, Ulf |
| contents | Large Language Models (LLMs) rely heavily on Key-Value (KV) caching to minimize inference latency. However, standard KV caches are context-dependent: reusing a cached document in a new context requires recomputing KV states to account for shifts in attention distribution. Existing solutions such as CacheBlend, EPIC, and SAM-KV mitigate this issue by selectively recomputing a subset of tokens; however, they still incur non-negligible computational overhead (FLOPs) and increased Time-to-First-Token (TTFT) latency. In this paper, we propose KV Packet, a recomputation-free cache reuse framework that treats cached documents as immutable ``packets'' wrapped in light-weight trainable soft-token adapters, which are trained via self-supervised distillation to bridge context discontinuities. Experiments on Llama-3.1 and Qwen2.5 demonstrate that the proposed KV Packet method achieves near-zero FLOPs and lower TTFT than recomputation-based baselines, while retaining F1 scores comparable to those of the full recomputation baseline. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2604_13226 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | KV Packet: Recomputation-Free Context-Independent KV Caching for LLMs Chen, Chuangtao Zhang, Grace Li Yin, Xunzhao Zhuo, Cheng Li, Bing Schlichtmann, Ulf Machine Learning Artificial Intelligence Large Language Models (LLMs) rely heavily on Key-Value (KV) caching to minimize inference latency. However, standard KV caches are context-dependent: reusing a cached document in a new context requires recomputing KV states to account for shifts in attention distribution. Existing solutions such as CacheBlend, EPIC, and SAM-KV mitigate this issue by selectively recomputing a subset of tokens; however, they still incur non-negligible computational overhead (FLOPs) and increased Time-to-First-Token (TTFT) latency. In this paper, we propose KV Packet, a recomputation-free cache reuse framework that treats cached documents as immutable ``packets'' wrapped in light-weight trainable soft-token adapters, which are trained via self-supervised distillation to bridge context discontinuities. Experiments on Llama-3.1 and Qwen2.5 demonstrate that the proposed KV Packet method achieves near-zero FLOPs and lower TTFT than recomputation-based baselines, while retaining F1 scores comparable to those of the full recomputation baseline. |
| title | KV Packet: Recomputation-Free Context-Independent KV Caching for LLMs |
| topic | Machine Learning Artificial Intelligence |
| url | https://arxiv.org/abs/2604.13226 |