AI Paper Atlas

Research map / Reinforcement learning / RL from feedback

Escaping Policy Contraction: Contraction-Aware PPO (CaPPO) for Stable Language Model Fine-Tuning

Dun Yuan, Di Wu, Xue Liu

ICLR 2026 · Reinforcement learning › RL from feedback · approach: GRPO stability and variants

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
replace PPO with CaPPO (Pareto multi-objective updates plus entropy scheduling)PPO · HH-RLHF, Qwen2.5-14Bwin rate vs SFT (%): 65.1 ± 0.7 -> 69.0 ± 0.5Improved
replace PPO with CaPPOPPO · UltraFeedback, Llama-3-8B-Instructwin rate vs SFT (%): 62.4 ± 0.8 -> 66.9 ± 0.7Improved
add entropy scheduling to PPO (ablation)PPO · HH-RLHF, Qwen2-7Bwin rate vs SFT (%): 62.8 ± 0.7 -> 64.3 ± 0.7Improved
add entropy scheduling to PPO (ablation)PPO · ShareGPT-1K, Llama3-8Bentropy (nats/token): 3.42 -> 3.76Improved
add entropy scheduling to PPO (ablation)PPO · ShareGPT-1K, Llama3-8BRepeat-2 (%): 24.5 -> 18.0Improved
add entropy scheduling to PPO (ablation)PPO · ShareGPT-1KSRR@tau: 0.37 -> 0.58Improved
use adaptive instead of fixed entropy coefficient in PPO (ablation)PPO · HH-RLHFSRR: 0.43 -> 0.59Improved
add Pareto multi-gradient update to adaptive-entropy PPO (ablation)PPO (adaptive beta) · HH-RLHFwin rate (%): 65.1 -> 67.8Improved
use Pareto update instead of scalarized entropy weight (ablation)PPO (lambda_H=0.3) · Summarize-from-Feedbackwin rate (%): 58.0 -> 61.9Improved
raise scalarized entropy weight in PPO from 0.0 to 0.1 (ablation)PPO · Summarize-from-Feedbackwin rate (%): 57.2 -> 58.3Improved
replace PPO with CaPPOPPO · RLHF macro-average, Qwen2-7BSRR: 0.55 -> 0.82Improved
use off-policy methods (RRHF) instead of on-policy PPOPPO · RLHF macro-average, Qwen2-7Bwin rate vs SFT (%): 67.4 -> 64.7 win, SRR 0.55 -> 0.92Mixed

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy