现代检索代理公开了许多配置选择 - LLM, 检索器, 文档数量, 跳数, 和合成策略 - 每个都影响答案质量和服务成本。如今, 这些管道通常针对每个工作负载, 手动调整一次，从而导致大量的每个查询优化尚未开发。我们在给定自然语言查询和准确性或预算目标,的情况下制定问题:，从预定义的管道目录中选择在推理时最小化成本或最大化准确性的配置。我们建议 **BRANE**, 它使用 LLM 将每个查询转换为特定于工作负载的特征,，然后训练一个轻量级的每个配置预测器，该预测器估计管道是否会正确回答查询。在推理时, **BRANE** 选择最大化预测正确性的配置，该配置受到成本, 的惩罚，从而无需重新训练即可实现可调整的成本质量权衡。在 MuSiQue, BrowseComp-Plus, 和 FinanceBench, **BRANE** 始终如一地推动成本质量 Pareto 前沿, 与最佳固定配置的 精度相匹配，成本降低高达 89%,，并且优于 LLM 路由, 基于规则的, 和微调的 Qwen3-4B 基线。这些结果表明，完整检索管道的每个查询配置是静态工作负载级别调整的实用替代方案。
Modern retrieval agents expose many configuration choices -- LLM, retriever, number of documents, number of hops, and synthesis strategy -- each shaping both answer quality and serving cost. Today, these pipelines are typically hand-tuned once per workload, leaving substantial per-query optimization untapped. We formulate the problem: given a natural-language query and either an accuracy or a budget target, select from a predefined pipeline catalog the configuration that minimizes cost or maximizes accuracy at inference time. We propose **BRANE**, which uses an LLM to convert each query into workload-specific characteristics, then trains a lightweight per-configuration predictor that estimates whether the pipeline will answer the query correctly. At inference time, **BRANE** selects the configuration that maximizes predicted correctness penalized by cost, exposing a tunable cost-quality tradeoff without retraining. Across MuSiQue, BrowseComp-Plus, and FinanceBench, **BRANE** consistently pushes the cost-quality Pareto frontier, matches the best fixed configuration的 accuracy at up to 89% lower cost, and outperforms LLM-routing, rule-based, and fine-tuned Qwen3-4B baselines. These results show that per-query configuration of the full retrieval pipeline is a practical alternative to static workload-level tuning.
科目: 人工智能 (cs.AI); 系统与控制 (eess.SY)
Subjects: Artificial Intelligence (cs.AI); Systems and Control (eess.SY)