TRINITY: An Evolved LLM Coordinator

TL;DR

TRINITY introduces a lightweight, evolved coordinator that orchestrates multiple LLMs through role assignment, significantly improving performance across various tasks without weight-merging limitations.

Contribution

The paper presents a novel evolutionary strategy-optimized coordinator model that effectively manages diverse LLMs, enabling superior task performance and robustness.

Findings

Outperforms individual models and existing methods on coding, math, reasoning, and domain tasks.

Achieves state-of-the-art 86.2% on LiveCodeBench.

Coordinator's hidden states and high-dimensional optimization are key to performance.

Abstract

Combining diverse foundation models is promising, but weight-merging is limited by mismatched architectures and closed APIs. Trinity addresses this with a lightweight coordinator that orchestrates collaboration among large language models (LLMs). The coordinator, comprising a compact language model (approximately $0.6$B parameters) and a lightweight head (approximately $10$K parameters), is optimized with an evolutionary strategy for efficient and adaptive delegation. Trinity processes queries over multiple turns, where at each turn the coordinator assigns one of three roles (Thinker, Worker, or Verifier) to a selected LLM, effectively offloading complex skill acquisition from the coordinator itself. Experiments show that Trinity consistently outperforms individual models and existing methods across coding, math, reasoning, and domain knowledge tasks, and generalizes robustly to out-of-distribution tasks. On standard benchmarks, Trinity achieves state-of-the-art results, including a score of 86.2% on LiveCodeBench. Theoretical and empirical analyses identify two main factors behind this performance: (1) the coordinator's hidden-state representations provide rich contextualization of inputs, and (2) under high dimensionality and strict budget constraints, the separable Covariance Matrix Adaptation Evolution Strategy offers advantages over reinforcement learning, imitation learning, and random search by exploiting potential block-epsilon-separability.