Latent Action Reparameterization for Efficient Agent Inference

TL;DR

The paper introduces Latent Action Reparameterization (LAR), a method that learns compact, semantic latent actions to reduce decision horizons and improve inference efficiency in large language model agents.

Contribution

LAR is a novel framework that learns latent action spaces from agent trajectories, enabling more efficient decision-making over shorter horizons without sacrificing expressiveness.

Findings

LAR significantly reduces effective action horizon across benchmarks.

LAR decreases action tokens and inference time while maintaining or improving success rates.

LAR outperforms traditional macro or hierarchical approaches in efficiency.

Abstract

Large language model (LLM) agents often rely on long sequences of low-level textual actions, resulting in large effective decision horizons and high inference cost. While prior work has focused on improving inference efficiency through system-level optimizations or prompt engineering, we argue that a key bottleneck lies in the representation of the action space itself. We propose Latent Action Reparameterization (LAR), a framework that learns a compact latent action space in which each latent action corresponds to a multi-step semantic behavior. By reparameterizing agent actions into latent units, LAR enables decision making over a shorter effective horizon while preserving the expressiveness of the original action space. Unlike hand-crafted macros or hierarchical controllers, latent actions are learned from agent trajectories and integrated directly into the model, allowing both planning and execution to operate over abstract action representations. Across a range of LLM-based agent benchmarks, LAR significantly reduces the effective action horizon and improves inference efficiency under fixed compute budgets. As a consequence, our approach achieves substantial reductions in action tokens and corresponding wall-clock inference time, while maintaining or improving task success rates. These results suggest that action representation learning is a critical and underexplored factor in scaling efficient LLM agent inference, complementary to advances in model architecture and hardware.