arXiv cs.AI· Sirui Zhang, Fangyun Wei, Dong Chen, Jinjing Zhao, Yitong Wang, Yunuo Chen, Baining Guo, Zheng-Jun Zha·· 3 天前AI 评分27
APT：面向文本生成图像的自反馈智能体后训练框架
APT: Feedback-Driven Agentic Post-Training for Text-to-Image Generation
AI 导读
APT 是一个全自动后训练框架，通过协调三个智能体构建评测集、诊断模型能力短板，并按自适应提示词分布生成强化学习数据，让模型当前表现决定下一轮训练任务，形成递归自我改进式反馈闭环。针对无用户指定评测场景的通用设置，该研究还开发了可生成广泛多样评测集的评测工具。实验表明 APT 在多个基座模型、benchmark 和生成架构上均适用，具备模型无关、benchmark 无关、架构无关且全自动的特性。
正文 · 原文
Abstract:Post-training is crucial for improving text-to-image models, yet existing pipelines typically rely on manually curated training data and fixed training distributions. We introduce APT (Agentic Post-Training), a fully automated framework that adaptively post-trains a base model according to its evolving capability profile. APT coordinates three agents to construct evaluation sets from user requirements, diagnose model-specific strengths and failure modes, and generate reinforcement learning data with an adaptive prompt distribution that emphasizes underperforming capabilities. By repeatedly evaluating, diagnosing, and updating the generator, APT lets the current model's behavior shape its next training tasks, forming a feedback loop inspired by recursive self-improvement. For general-purpose settings without user-specified evaluation scenarios, we further develop an evaluation harness that enables broad and diverse evaluation set generation. Experiments across multiple base models, benchmarks, and generation architectures demonstrate that APT is model-agnostic, benchmark-agnostic, architecture-agnostic, and fully automated.
| Subjects: | Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2610.18097 [cs.CV] |
| (or arXiv:2610.18097v1 [cs.CV] for this version) | |
| https://doi.org/10.48550/arXiv.2610.18097 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Sirui Zhang [view email]
[v1]
Tue, 13 Oct 2026 10:14:23 UTC (9,678 KB)
来源：arXiv cs.AI · arxiv.org