AI Paper Atlas

Research map / Reinforcement learning / RL from feedback

rePIRL: Learn PRM with Inverse RL for LLM Reasoning

Xian Wu, Kaijie Zhu, Ying Zhang, Lun Wang, Wenbo Guo

ICML 2026 · Reinforcement learning › RL from feedback · approach: RL for LLM reasoning · 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
learn PRM with inverse RL (rePIRL) instead of vanilla outcome-reward RLOOQwen2.5-3B-Instruct · Math averageaccuracy: 31.7 -> 33.5Improved
learn PRM with inverse RL (rePIRL) instead of vanilla RLOOQwen3-4B-Base · Coding averageaccuracy: 35.5 -> 39.8Improved
use rePIRL PRM versus PRIME online PRMQwen3-4B-Base · Math averageaccuracy: 39.3 -> 41.6Improved
add SFT warm-up on expert trajectories before RLOOQwen2.5-3B-Instruct · Math averageaccuracy: 31.7 -> 25.7 (RLOO vs SFT+RLOO)Hurt
offline methods (BC, KTO) instead of online RLQwen2.5-3B-Instruct · Math averageaccuracy: 28.4 -> 20.6 (BC)Hurt
use rePIRL PRM alone (no outcome reward) to train policy from scratch on MATHQwen3-4B-Base · Test-time training avgaccuracy: 42.3 (RLOO) vs 42.1 (rePIRL)No clear effect
use PRM to select responses instead of majority votingQwen3-4B-Base · AIME-2024accuracy: 0.200 -> 0.230 (majority voting vs rePIRL at 64 samples)Improved
train on hard problems with rePIRL PRM instead of outcome rewardQwen3-4B-Base · Hard MATH subsetaccuracyImproved
remove outcome reward, train with PRM only (ablation)rePIRL · MATH-500accuracy: 46.0 -> 59.6 (vs base model)Improved
add corrected policy trajectories as additional expert data for reward model (ablation)rePIRL · training stabilitystability and performanceImproved
rePIRL vs RLOO on MATH500 with Qwen2.5-3B-InstructQwen2.5-3B-Instruct · MATH500pass@1: 63.6 -> 62.4No clear effect
rePIRL vs PRIME on LiveCodeBenchQwen3-4B-Base · LiveCodeBenchpass@1: 19.0 -> 27.5Improved

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