Generalizing Test-time Compute-optimal Scaling as an Optimizable Graph

TL;DR

This paper introduces a novel framework for optimizing large language model collaboration architectures at test time, using probabilistic graph optimization and LLM-agent feedback to improve inference efficiency and accuracy.

Contribution

It formalizes the problem as a probabilistic graph optimization and proposes Agent-REINFORCE, a new LLM-agent-based method for efficiently searching for optimal collaboration graphs.

Findings

Agent-REINFORCE outperforms baselines in search efficiency.

It effectively balances accuracy and inference latency.

The method identifies superior model collaboration architectures.

Abstract

Test-Time Scaling (TTS) improves large language models (LLMs) by allocating additional computation during inference, typically through parallel, sequential, or hybrid scaling. However, prior studies often assume fixed collaboration architectures (e.g., topologies) and single-model usage, overlooking that optimal architectures and model combinations can vary across tasks. Therefore, we study the novel problem of searching for compute-optimal model combinations and architectures in TTS under a fixed budget. We formalize it as a multi-LLM collaboration graph, where nodes encode roles and LLM model assignments, and edges capture information flow. This problem is challenging because (i) the combinatorial search space is prohibitively large, and (ii) task-specific requirements demand tailored designs. To address these, we reformulate the problem as probabilistic graph optimization and, through pilot experiments, derive three empirical insights into TTS collaboration graphs. Guided by these insights, we propose Agent-REINFORCE, an LLM-agent-augmented framework that mirrors the REINFORCE pipeline by mapping sampling-gradient-update to sampling-feedback-update, where feedback serves as a textual gradient to update the probabilistic graph and efficiently search for optimal multi-LLM collaboration graphs. Experiments show that Agent-REINFORCE outperforms both traditional and LLM-based baselines in sample efficiency and search performance, and effectively identifies optimal graphs under joint objectives of accuracy and inference latency.