← 首页|学术|TTPO: Test-Time Policy Optimization
cs.CL · 2608.27448 · 2026-08-27

TTPO: Test-Time Policy Optimization

Aozhe Wang, Zhengxi Lu, Jianze Wang, Shangke Lv, Ying Liu, Weiming Lu, Jun Xiao, Yueting Zhuang, Hua Yang, Qianglong Chen, Yongliang Shen
Test-Time TrainingRL训练推理
💬 不用标注数据做测试时训练:用「投票分歧」这个不对称信号同时做蒸馏和惩罚,让 1.7B 模型在测试时自我提升 7 个点。

🎯 背景

RL/OPSD 等后训练方法都依赖 ground-truth 标签,无法做测试时训练(TTT)。用多数投票伪标签替代标签很自然,但一旦投票本身错了会带偏整个教师信号。

🔬 方法

TTPO 发现一个不对称现象:与伪标签「不一致」的 rollout 无论投票对错基本都是错的。据此设计不对称目标——同意伪标签的 rollout 用 OPSD 蒸馏,不同意的用 Grouped RL 惩罚,并在 token 级别做进一步筛选(蒸馏跳过已收敛位置,RL 只惩罚高置信度错误)。

📊 结果与意义

在五个竞赛级数学基准上,无标签的 TTPO 打平了有标签监督的 OPSD;Qwen3-1.7B 测试时训练从 38.0% 提升到 45.2%,跨任务泛化也表现稳健。
▶ 原文摘要 Abstract
Recent prominent post-training methods, such as Reinforcement Learning (RL) and On-Policy Self-Distillation (OPSD), have driven rapid progress in mathematical reasoning for large language models, yet their reliance on ground-truth labels precludes test-time training (TTT). Replacing ground truth with majority-vote pseudo-labels is a natural alternative, yet it is fragile: an incorrect vote corrupts the teacher and misleads every token. We observe that this failure mode is asymmetric: rollouts that disagree with the pseudo-label are typically wrong regardless of whether the vote itself is correct. Building on this observation, we propose Test-Time Policy Optimization (TTPO), an asymmetric objective that distills agreeing rollouts via OPSD and penalizes disagreeing rollouts with Grouped RL. Token-level selection further refines both branches: distillation down-weights already-converged positions, while RL penalizes only confident errors. Both updates remain well-grounded even under frequent pseudo-label errors, and majority-vote routing yields tighter self-supervision as the model improves. Without any labels, TTPO matches label-supervised OPSD on five competition-level benchmarks, raises Qwen3-1.7B from 38.0% to 45.2% in TTT, yields +25.2% to +36.4% without thinking, and shows strong cross-task generalization.
来源:arXiv:2608.27448 · 精读基于摘要与 arXiv HTML/abs 页信息生成,未解析 PDF 全文