AI Paper Atlas

Research map / Reinforcement learning / RL from feedback

PS-PPO : Prefix-Sampling PPO for Critic-Free RLHF

Doo Hwan Hwang, Kee-Eung Kim

ICML 2026 · Reinforcement learning › RL from feedback · approach: RL fine-tuning of generative models · 0 citations (OpenAlex)

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
optimize the monotone cutoff distribution instead of uniform (ablation)PS-PPO, Qwen2.5-Math-7B · AIME25pass@1 accuracy (%): 9.0 -> 13.3Improved
optimize cutoff distribution instead of heuristic post-processing (ablation)PS-PPO, Qwen2.5-Math-7B · AIME25pass@1 accuracy (%): 10.0 -> 13.3Improved
optimize cutoff instead of fixed-length prefix (ablation)PS-PPO, Qwen2.5-Math-7B · AIME24pass@1 accuracy (%): 10.0 -> 16.7Improved
use exponential time prior instead of uniform cutoff (ablation)PS-PPO · math benchmarkspass@1 accuracy (%)No clear effect
raise budget B from 64 to 128 (ablation)PS-PPO, Qwen2.5-Math-7B · six math benchmarks averageaverage pass@1 (%): 34.0 -> 37.9Improved
raise budget B from 128 to 512 (ablation)PS-PPO, Qwen2.5-Math-7B · six math benchmarks averageaverage pass@1 (%): 37.9 -> 37.8No clear effect
raise budget B from 128 to 512: time (ablation)PS-PPO · training time per stepseconds: 1.77 -> 2.56Hurt
raise rollouts K from 2 to 16 (ablation)PS-PPO, Qwen2.5-Math-7B · six math benchmarks averageaverage pass@1 (%): 34.2 -> 38.5Improved
raise rollouts K from 8 to 16 (ablation)PS-PPO · six math benchmarks averageaverage pass@1 (%): 37.9 -> 38.5No clear effect
use PS-PPO prefix truncation instead of DAPO full-sequenceQwen2.5-Math-7B · training time per stepseconds: 3.23 -> 1.77Improved
PS-PPO instead of DAPO at Tmax=4096Qwen2.5-Math-7B · update-stage timespeedup: 3.3x speedup over DAPOImproved
PS-PPO instead of DAPO at Tmax=2048Qwen2.5-Math-7B · average of MATH500, AIME24, AIME25pass@1 accuracy (%): 42.1 -> 42.1No clear effect

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