KV Cache 能否跨模型版本复用？DeepSeek-V4-Flash 案例研究
Can KV Cache Be Reused Across Model Updates? A DeepSeek-V4-Flash Case Study
做推理服务优化的人该看看这篇：升级模型版本后旧 cache 别急着扔，在长前缀 QA 场景下能拿到 2 到 5 倍的首 token 加速。论文用 DeepSeek-V4-Flash 做了完整实验，还给出了 FCD 这个新指标来量化复用后行为偏移，比单纯看准确率更细。
arXiv 论文（编号 2611.14051）研究了 KV cache 在语言模型版本更新之间的直接复用问题，涵盖四次模型升级，并重点分析了 DeepSeek-V4-Flash 到 Flash-0731 的案例。核心发现是：旧版本的 cache 能保留有用的任务信息，但并不能忠实还原新模型的行为。
在三个较长前缀的 QA 基准上，利用已有的源 cache，复用可实现中位数 2.10 到 5.08 倍的 time-to-first-token 加速。检索精度在不同版本更新之间差异很大，从接近原生性能到大幅损失都有。论文提出 Functional Cache Divergence 指标，通过比较复用 cache 与原生 cache 下的 next-token 分布来刻画行为差异：在自然文本续写任务中复用效果较好，而在单 token 检索任务中，前缀越长兼容性越差。
研究还发现，即使第一个生成 token 一致，也不保证答案内容和停止行为一致。这意味着评估跨版本 KV cache 复用需要把任务效用和对原生推理的保真度作为两个独立标准，对推理服务的成本优化有直接参考价值。
Can KV Cache Be Reused Across Model Updates? A DeepSeek-V4-Flash Case Study
Reusing key--value (KV) caches across language model updates could avoid repeated computation of shared context, but changes in model weights raise a fundamental question about what these caches preserve. We investigate direct cache reuse across four model updates, with a detailed study of DeepSeek-V4-Flash to Flash-0731. Our central finding is that caches can retain useful task information without faithfully reproducing the updated model's behavior. Retrieval accuracy varies substantially across updates, ranging from near-native performance to large losses. With source caches already available, reuse achieves median time-to-first-token speedups of $2.10$--$5.08\times$ on three longer-prefix QA benchmarks. To characterize the behavioral differences behind these outcomes, we introduce Functional Cache Divergence, which compares next-token distributions under reused and native caches. Direct KV cache reuse better preserves native model behavior in natural-text continuation, whereas compatibility degrades with increasing prefix length in single-token retrieval tasks. Answer-margin analyses help explain how correct retrieval decisions survive substantial internal state changes. Yet even agreement on the first generated token does not ensure agreement in answer content or stopping behavior. These findings establish task utility and fidelity to native inference as distinct criteria for evaluating cross-version KV cache reuse.
This report is categorized under AITOP's "Papers" desk, focusing on China's AI industry and domestic foundation model ecosystem. Within our cross-regional analytical framework, it directly maps to AI developer toolchains, coding agents, and engineering workflows. Key entities and technical topics under continuous tracking include DeepSeek-V4-Flash, KV Cache, 推理优化, RAG.
Primary source verification: published directly by arXiv: DeepSeek with first-party authority.