压缩键值缓存时，答案正确不等于推理可信。
Does Accuracy Equal Evidence? Reasoning Faithfulness under KV Cache Compression
- 用固定推理轨迹测试压缩对推理过程的影响
- 部分方法保答案却丢推理支撑，出现答案-证据鸿沟
- 量化方法比淘汰令牌更保推理完整性，适合高可靠性场景
KV缓存压缩通常通过最终答案准确率评估，隐含假设是答案正确即推理过程也保留。我们检验该假设在大模型推理中的有效性，发现其可能失效：压缩下正确答案与可见推理链的有效性可被不同速率破坏。通过受控的固定轨迹重放协议，固定推理内容并隔离压缩是否保留可用信息。在三个模型上评估十种令牌淘汰压缩方法及一种量化方法，涵盖数学推理、科学问答、临床计算和长上下文检索任务。测量最终准确率、答案链一致性与扰动忠实度。结果显示，多数令牌淘汰方法可保持较高答案准确率，但显著降低推理链支持力或扰动忠实度。我们称之为‘答案-证据差距’。而一种保持覆盖率的量化控制方法受影响较小，表明失败根源并非内存减少本身，而是丢失推理轨迹中部分关键信息。代码已开源。
原文摘要 · Abstract (English)
KV cache compression is commonly evaluated by final-answer accuracy, implicitly assuming that preserving the answer also preserves the reasoning that supports it. We test this assumption for large reasoning models and show that it can fail: under compression, correct answers and the validity of their visible supporting rationales can be preserved at different rates. We study this failure with a controlled fixed-trace replay protocol, which holds reasoning content fixed and isolates whether compression preserves usable information from an already available trace. We evaluate ten token-eviction KV compression methods and one quantization method on three models across mathematical reasoning, scientific QA, clinical calculation, and long-context retrieval. We measure final accuracy, answer-chain consistency, and perturbation faithfulness. Across tasks, token-eviction methods can preserve competitive final-answer accuracy while substantially degrading chain support or perturbation faithfulness. We call this the answer-evidence gap. A coverage-preserving quantization control is substantially less affected, suggesting that the failure is tied less to KV memory reduction itself than to losing access to parts of the reasoning trace. Code is available at https://github.com/famous-blue-raincoat/Safe_KV_Compress.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。