PulseAugur
中
实时 14:23:15
English(EN) CurateEvo: Data-Curation Evolving for Agentic Post-Training

新框架CurateEvo增强LLM Agent训练后数据策展 · 追踪2个来源

作者 PulseAugur 编辑部 · [4 个来源] ·

研究人员开发了CurateEvo，一个用于动态演进数据策展策略的新框架，以改进大型语言模型（LLM）Agent的训练后阶段。这种由失败驱动的方法通过分析失败的轨迹来迭代地优化策展方法，从而为微调和强化学习提供更有效和高效的数据准备。在ACEBench-Agent和tau^2-Bench等基准上的实验表明，CurateEvo的性能持续优于现有的策展技术，提高了Agent的性能并降低了开销。 AI

影响 这项研究介绍了一种更有效和高效的方法来为LLM Agent准备数据，有可能提高它们在复杂、长时域任务上的性能。

排序理由 该集群包含两个相同的arXiv预印本，详细介绍了LLM Agent的新数据策展框架。

在 arXiv cs.CL 阅读 →

AI 生成摘要 · Google Gemini · 来自 4 个来源。 我们如何撰写摘要 →

新框架CurateEvo增强LLM Agent训练后数据策展 · 追踪2个来源

报道来源 [4]

  1. arXiv cs.CL TIER_1 English(EN) · Dingzirui Wang, Xuanliang Zhang, Keyan Xu, Qingfu Zhu, Wanxiang Che ·

    CurateEvo：Agentic训练后数据策展的演进

    arXiv:2607.06140v1 Announce Type: new 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 preproce…

  2. arXiv cs.CL TIER_1 English(EN) · Wanxiang Che ·

    CurateEvo：Agentic训练后数据策展的演进

    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…

  3. Hugging Face Daily Papers TIER_1 English(EN) ·

    CurateEvo：Agentic 训练后数据策展的演进

    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…

  4. arXiv cs.AI TIER_1 English(EN) · Junze Ye, Jiayi Cheng, Miao Lu, Michal Mankowski, Jose Blanchet, Mohsen Bayati ·

    少量教师步骤效果显著：代理训练后成本效益高的在线策略数据增强

    arXiv:2607.04574v1 Announce Type: cross Abstract: For LLM agents, supervised fine-tuning is not only about teacher labels' quality, but also about which interaction contexts those labels condition on. Pure behavioral cloning uses full teacher demonstrations, creating a mismatch b…

相关实体

相关话题