解决对话系统历史依赖过强问题，让模型更灵活适应意图变化。
Overcoming State Inertia: Minimally Invasive Temporal Alignment for Evolving Contexts
- 通过冲突感知优化与时间注意力偏置，分离状态更新与语言建模。
- 在多轮对话和新基准上保持结构连贯性，困惑度仅微升。
- 支持低延迟静态模式与高精度动态模式，适合不同场景需求。
长对话系统存在状态惯性问题，模型过度关注历史而难以适应意图演变。我们发现标准对齐方法如DPO及近期长上下文优化技术在解决此问题时会引发严重上下文对齐代价——显著提升困惑度，破坏预训练先验。为此，提出DZ-TiDPO框架，融合冲突感知优化（训练阶段）与结构化时间注意力偏置，有效解耦状态更新与通用语言建模。在Multi-Session Chat和新构建的Inertia Challenge（IC-Bench）数据集上的实验表明，该框架在保持结构连贯性的同时，有效缓解跨轮次冲突。关键优势在于支持双推理策略：低延迟静态模式用于通用鲁棒性，精度导向动态模式用于细粒度语义冲突处理。规模分析揭示容量-稳定性权衡，证实中等规模模型（7B）可高效内化时间对齐能力。代码与数据已公开于https://github.com/lyj20071013/DZ-TiDPO。
原文摘要 · Abstract (English)
Long-context dialogue systems suffer from state inertia, where models over-attend to history and fail to adapt to evolving intents. We demonstrate that standard alignment methods like DPO and even recent long-context optimization techniques struggle to resolve this without incurring a severe contextual alignment tax--a substantial perplexity surge caused by disrupting pre-trained priors. To address this, we propose DZ-TiDPO, a minimally invasive framework that synergizes conflict-aware optimization (during training) with a structural temporal attention bias. This design effectively decouples state updating from general linguistic modeling. Experiments on Multi-Session Chat and our new Inertia Challenge (IC-Bench) show DZ-TiDPO preserves structural coherence while resolving inter-turn conflicts. Crucially, our framework supports dual inference strategies: a negligible-latency static mode for general robustness and a precision-focused dynamic mode for micro-semantic conflicts. Furthermore, our scaling analysis reveals a capacity-stability trade-off, confirming that highly capable mid-sized models (7B) can efficiently internalize temporal alignment. Code and data are available at: https://github.com/lyj20071013/DZ-TiDPO.
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。