Multi-agent reasoning improves solutions by training at test time

Consolidating Reasoning with Test-Time Training

Computation and LanguageArtificial Intelligence

Summary

Solving problems with many helpers sometimes wastes the helpful steps they take along the way and only looks at their final answers. The authors created a way to learn from these helpful steps during problem solving itself, so the computer gets better as it works on a problem. They trained this approach using math problems and showed it helps get better answers, especially with more helpers and bigger models. This method is also good at fixing errors from bad individual tries and learns best when it can practice both supervised and reinforcement learning styles.

What this means in practice

  • For ai system engineers: Improve multi-agent AI systems by consolidating reasoning steps during problem solving to boost final solution accuracy.
  • For education technology developers: Develop smarter tutoring systems that learn from multiple solution attempts to better assist with STEM problem solving.

Authors

Haotian Wu, Zeming Chen, Hao Zhao, Antoine Bosselut

Abstract

Effective multi-agent collaboration requires finalizing solutions from reasoning that is distributed across the individual attempts of many agents. However, useful intermediate reasoning can remain unexploited when collaboration is limited to voting on conclusions reached among individual agents, or aggregating long, noisy collections of reasoning trajectories in context. In this work, we propose TTT-Consolidate, a framework that uses multi-agent reasoning trajectories as adaptation data to improve a model's subsequent reasoning. TTT-Consolidate jointly consolidates trajectories into parametric representations through test-time training, then generates a new solution by conditioning on the consolidated information. We meta-learn how to consolidate using bi-level optimization at training time, where an inner loop encodes reasoning trajectories into a lightweight LoRA adapter, and an outer loop learns meta-parameters for the inner loop such that the consolidation is helpful for the model to synthesize a final solution. As no large-scale corpus of diverse reasoning trajectories exists for training aggregation methods, we construct the OpenParallelThinking corpus composed of 12,810 math problems paired with a mixture of trajectories from six LLMs. Evaluated on challenging math and STEM benchmarks, TTT-Consolidate outperforms the best aggregation baselines by up to 20 percentage points, even under matched compute and runtime settings, with the gap widening at larger model scales. TTT-Consolidate also recovers from unsuccessful individual trajectories more robustly, an ability that improves as the number of available (even incorrect) trajectories increases. Finally, we demonstrate that TTT-Consolidate benefits from both SFT and RL-based outer loop objectives. We will release TTT-Consolidate and OpenParallelThinking for use in future work on reasoning aggregation.