A Game Theoretic Free Energy Analysis of Higher Order Synergy in Attention Heads of Large Language Models

TL;DR

This paper introduces a game-theoretic framework to analyze attention heads in large language models, revealing redundancies and enabling effective pruning with minimal performance impact.

Contribution

It applies the Game Theoretic Free Energy Principle to interpret attention heads as bounded rational agents, providing insights into their interactions and redundancy.

Findings

Triple dividends are consistently negative, indicating higher order redundancy.

Pruning 20% of heads reduces FLOPs by 18% and increases throughput by 22%.

Minimal performance loss observed after pruning, with perplexity increasing modestly.

Abstract

Large language models rely on multihead attention, but interactions among heads remain poorly understood. We apply the Game Theoretic Free Energy Principle (GTFEP): a framework casting multiagent systems as distributed variational inference to analyze attention heads as bounded rational agents. According to GTFEP, each head minimizes its variational free energy, and collective behavior follows a Gibbs distribution over coalition structures whose energy is decomposed into Harsanyi dividends. Using a tractable approximation (uniform prior, deterministic dynamics), coalition free energy reduces to joint Shannon entropy of discretized head outputs (argmax key index). Pairwise dividends become mutual information (nonnegative), while triple dividends correspond to interaction information and can be negative. On BERT, GPT2, and Llama with GSM8K, triple dividends are consistently negative, revealing higher order redundancy. The Nash FEP correspondence guarantees that stationary points of collective free energy are epsilon Nash equilibria; thus, heads with negligible contribution can be pruned with minimal performance loss. Pruning heads with low marginal contribution reduces computational cost with minimal performance loss: for example, pruning 20% of heads in GPT2 reduces FLOPs by 18%, increases throughput by 22%, and raises perplexity only modestly (from 28.4 to 33.4 on GSM8K). Our work shows GTFEP provides a principled foundation for analyzing and optimizing transformer architectures.