workload-management

A paper proposing a latency-aware LLM query router that jointly optimizes latency, accuracy, and cost using a lightweight latency estimator, achieving up to 40% improvement in accuracy–cost utility while maintaining comparable latencies.