Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This paper introduces negative early exit and adaptive boosting to optimize Monte Carlo Tree Search (MCTS) for test-time compute scaling (TTCS) in large language models. Negative early exit prunes unproductive MCTS trajectories, while adaptive boosting reallocates computation to reduce resource contention. The integrated approach, implemented within vLLM, significantly reduces p99 latency and improves throughput without sacrificing reasoning accuracy.
MCTS gets a speed boost: pruning unproductive search trajectories and adaptively reallocating compute slashes p99 latency without hurting reasoning.
Monte Carlo Tree Search (MCTS) is an effective test-time compute scaling (TTCS) method for improving the reasoning performance of large language models, but its highly variable execution time leads to severe long-tail latency in practice. Existing optimizations such as positive early exit, reduce latency in favorable cases but are less effective when search continues without meaningful progress. We introduce {\it negative early exit}, which prunes unproductive MCTS trajectories, and an {\it adaptive boosting mechanism} that reallocates reclaimed computation to reduce resource contention among concurrent searches. Integrated into vLLM, these techniques substantially reduce p99 end-to-end latency while improving throughput and maintaining reasoning accuracy.