CurateEvo: Data-Curation Evolving for Agentic Post-Training

2026-07-07

AI Summary

This work addresses the limitation of existing agent post-training methods, which treat data curation as a static preprocessing step and neglect dynamic feedback from downstream task failures. To overcome this, we propose CurateEvo, a novel framework that introduces, for the first time, a failure-driven dynamic evolution mechanism. In CurateEvo, curation policies are represented as executable code and iteratively rewritten based on failure trajectories observed on a validation set. This process automatically generates supervision signals for fine-tuning, reinforcement learning data, and a reasoning memory bank, while incorporating cost-aware pruning to enhance computational efficiency. Evaluated on ACEBench-Agent, BFCL-V4, and τ²-Bench, our approach achieves average improvements of 3.2 and 2.7 points over state-of-the-art methods, while significantly reducing data curation overhead.

Abstract

Large language model (LLM) agents require post-training methods that can improve long-horizon decision making from environment feedback. However, existing agentic post-training pipelines often treat data curation as a fixed preprocessing step, focusing mainly on data augmentation while neglecting filtering, refinement, and adaptation to downstream failures. We propose CurateEvo, a failure-driven dynamic evolution framework for agentic post-training data curation. CurateEvo represents the curation strategy as executable code and iteratively rewrites it using failed trajectories from a held-out development set. At each epoch, the evolved strategy transforms a fixed raw corpus into supervised fine-tuning data, reinforcement learning data, and an inference-time memory bank. The evolution process first improves effectiveness by diagnosing recurring failure modes and augmenting, filtering, or refining data accordingly, and then improves efficiency by pruning redundant or low-utility training turns under a cost-aware objective. Experiments on ACEBench-Agent, BFCL-V4, and τ^2-Bench under both labeled and wild-data settings show that CurateEvo consistently outperforms prior curation methods, improving average scores by 3.2 and 2.7 points, respectively. Further analyses demonstrate that CurateEvo is compatible with different post-training recipes and substantially reduces curation overhead.

Problem

• data curation
• agentic post-training
• failure-driven
• long-horizon decision making
• LLM agents

Innovation

• data curation
• agentic post-training
• failure-driven evolution
• dynamic strategy rewriting
• cost-aware pruning