Divide, Optimize, Merge: Fine-Grained LLM Agent Optimization at Scale
This work addresses scalability issues in LLM agent optimization for AI researchers and practitioners, offering an incremental improvement over existing methods.
The paper tackles the challenge of scaling LLM-based optimization for agent systems by proposing the Fine-Grained Optimization (FGO) framework, which divides tasks into subsets and merges optimized components, resulting in performance improvements of 1.6-8.6% and a 56.3% reduction in prompt token consumption across benchmarks.
LLM-based optimization has shown remarkable potential in enhancing agentic systems. However, the conventional approach of prompting LLM optimizer with the whole training trajectories on training dataset in a single pass becomes untenable as datasets grow, leading to context window overflow and degraded pattern recognition. To address these challenges, we propose Fine-Grained Optimization (FGO), a scalable framework that divides large optimization tasks into manageable subsets, performs targeted optimizations, and systematically combines optimized components through progressive merging. Evaluation across ALFWorld, LogisticsQA, and GAIA benchmarks demonstrate that FGO outperforms existing approaches by 1.6-8.6% while reducing average prompt token consumption by 56.3%. Our framework provides a practical solution for scaling up LLM-based optimization of increasingly sophisticated agent systems. Further analysis demonstrates that FGO achieves the most consistent performance gain in all training dataset sizes, showcasing its scalability and efficiency.