Beyond Static RAG: An Adaptive, Tri-Metric Routing Framework for Efficient Long-Context Inference on Commodity GPUs
The Tri-Metric Router is introduced, a deterministic, training-free policy that selects among Raw, Neural, and Lexical pipelines and achieves 0% OOM failures, on out-of-distribution holdouts, and improves over always-on lexical compression by 5.2 points without additional VRAM or training cost.