Agent-Omit: Adaptive Context Omission for Efficient LLM Agents

TL;DR

Agent-Omit introduces an adaptive omission framework for LLM agents, reducing redundant thoughts and observations to enhance efficiency without sacrificing effectiveness, validated across multiple benchmarks.

Contribution

The paper proposes a novel training framework and reinforcement learning approach enabling LLM agents to adaptively omit unnecessary information, improving efficiency and effectiveness.

Findings

Agent-Omit-8B achieves comparable performance to frontier LLM agents.

The method outperforms seven existing efficient LLM agent approaches.

The omission policy deviation is theoretically bounded by KL-divergence.

Abstract

Managing agent context (e.g., thought and observation) during multi-turn agent-environment interactions is an emerging strategy to improve agent efficiency. However, existing studies treat the entire interaction trajectories equally, overlooking the thought necessity and observation utility varies across turns. To this end, we first conduct quantitative investigations into how thought and observation affect agent effectiveness and efficiency. Based on our findings, we propose Agent-Omit, a unified training framework that empowers LLM agents to adaptively omit redundant thoughts and observations. Specifically, we first synthesize a small amount of cold-start data, including both single-turn and multi-turn omission scenarios, to fine-tune the agent for omission behaviors. Furthermore, we introduce an omit-aware agentic reinforcement learning approach, incorporating a dual sampling mechanism and a tailored omission reward to incentivize the agent's adaptive omission capability. Theoretically, we prove that the deviation of our omission policy is upper-bounded by KL-divergence. Experimental results on five agent benchmarks show that our constructed Agent-Omit-8B could obtain performance comparable to seven frontier LLM agent, and achieve the best effectiveness-efficiency trade-off than seven efficient LLM agents methods. Our code and data are available at https://github.com/usail-hkust/Agent-Omit.