提出Pced框架，让检索增强生成实现多文档推理且不慢。
Parallel Context-of-Experts Decoding for Retrieval Augmented Generation
- 将证据融合从注意力机制移至解码阶段，用专家并行策略处理文档
- 在不共享注意力的前提下恢复跨文档推理能力，提升生成质量
- 无需训练，适合需要快速部署的RAG系统
检索增强生成面临权衡：将文档拼接成长提示可实现多文档推理，但引发预填充瓶颈；分别编码文档键值缓存虽提速，却破坏跨文档交互。我们提出无需训练的并行专家上下文解码（Pced）框架，将证据聚合从注意力机制转移到解码过程。Pced将检索到的文档视为独立的‘专家’，通过一种新颖的检索感知对比解码规则，根据模型先验权重专家输出概率。该方法在不构建跨文档共享注意力的情况下，恢复了跨文档推理能力。
原文摘要 · Abstract (English)
Retrieval Augmented Generation faces a trade-off: concatenating documents in a long prompt enables multi-document reasoning but creates prefill bottlenecks, while encoding document KV caches separately offers speed but breaks cross-document interaction. We propose Parallel Context-of-Experts Decoding (Pced), a training-free framework that shifts evidence aggregation from the attention mechanism to the decoding. Pced treats retrieved documents as isolated "experts", synchronizing their predictions via a novel retrieval-aware contrastive decoding rule that weighs expert logits against the model prior. This approach recovers cross-document reasoning capabilities without constructing a shared attention across documents.
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。