Adaptive Parallel Monte Carlo Tree Search for Efficient Test-time Compute Scaling

2026-04-01

AI Summary

This work addresses the high variability and long-tail latency in Monte Carlo Tree Search (MCTS) during test-time computation expansion, which stems from inefficient search trajectories and limits the effectiveness of existing optimizations when search progress stalls. The authors propose a negative early-exit mechanism that proactively prunes unproductive trajectories and integrates an adaptive boosting strategy to dynamically reallocate the freed computational resources, thereby mitigating resource contention among parallel searches. Implemented within the vLLM inference framework, this approach significantly reduces end-to-end p99 latency and improves system throughput while preserving the accuracy of large language model inference.

Technology Category

• Search and Optimization: Sampling/Simulation-based Search
• Machine Learning: Mixture of Experts (MoE)
• Planning, Routing, and Scheduling: Planning with Language Models

Application Category

• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• Search and Retrieval-Augmented AI: Search Tool Learning with LLM: Teaching LLMs to invoke search and make use of retrieved information
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

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.

Problem

• Monte Carlo Tree Search
• test-time compute scaling
• long-tail latency
• execution time variability
• large language models

Innovation

• negative early exit
• adaptive boosting
• Monte Carlo Tree Search
• test-time compute scaling
• latency optimization