arXiv 2610.26426 · cs.CL · 2026-10-19
HarnessRL: Co-Evolving Policies with a Just-in-Time Training Harness for Open-Ended RL
Xiaoqiang Wang, Ting-En Lin, Yuchuan Wu, Pinyi Zhang, Caixin Kang, Ling Li, Bingxue Qiu, Jieping Ye 等 10 人
Qwen-Character
一句话要点
HarnessRL通过动态演化任务采样、策略引导和评分标准,解决开放域强化学习中奖励失效问题,在多个领域任务上显著提升性能并减少训练步数。
HarnessRL dynamically evolves training interfaces to address reward failure in open-ended RL, improving performance and reducing training steps.
要点
- 开放域强化学习中,固定任务采样与评估导致奖励失效,表现为评分失真或崩溃,严重阻碍策略优化。
- HarnessRL引入动态演化机制,通过任务采样、策略引导和评分标准三个接口随策略演进而更新,利用历史数据判断失败原因并迭代修正。
- 在Rubric-Mix-10k数据集上,HarnessRL在4B和9B模型规模下比最强基线提升2.8至3.1点,训练步数几乎减半。
- 该方法通过即时反馈机制确保更新对当前策略有效,避免无效或有害的参数调整,但未明确说明其在复杂场景下的泛化能力。
English overview
HarnessRL addresses reward degradation in open-ended RL by co-evolving training interfaces with the policy. It dynamically updates task sampling, rollout guidance, and rubric criteria using within- and cross-task evidence. A just-in-time critic evaluates proposed changes on a frozen policy to ensure effectiveness. Evaluated on Rubric-Mix-10k, it outperforms baselines significantly and reduces training steps.
只根据 arXiv 官方摘要,由开源模型整理,经数字/专名自动核对(必须在摘要中出现),以论文原文为准。
阅读 arXiv 原文 PDF Hugging Face 讨论
主题与学科
同主题:强化学习
- Waypoint-OPD: Transferring RL-Induced Policy Shifts via Gated On-Policy Distillation ▲214
- Self-Retrospection Distillation: Turning Post-hoc Experiences into Prior Foresight ▲189
- LoGRA: Scaling LLM Reinforcement Learning with Low-Rank Gradient Sketches ▲152
- Sharpening Tax in Post-Training ▲111
- RULER: Instance-aware Rubric Rewards for SVG Generation ▲104
同主题:高效训练与推理
- MJev-Spatial: System One Spatial Intelligence via Native Attention Matching 2026-10-15 · ▲556
- The Teacher Is a Direction, Not a Destination: Extrapolating RL-Induced Representation Residuals in On-Policy Distillation 2026-09-29 · ▲511
- VideoJev: System One Model for Video Understanding and Interaction 2026-10-13 · ▲375
- Jeva: System One Models from First Principles 2026-10-13 · ▲371
- LEV: Thinking Fast and Deep with Looped Decision Models 2026-10-14 · ▲337
数据说明
元数据来自 arXiv API(已核实);点赞来自 Hugging Face 每日论文;引用来自 Semantic Scholar。最近刷新 2026-10-19。
要点由本站根据摘要自写,不转载摘要原文,请阅读原文。