AI Paper Atlas

Research map / Reinforcement learning / RL from feedback

GRPO is Secretly a Process Reward Model

Michael Sullivan, Alexander Koller

ICML 2026 · Reinforcement learning › RL from feedback · approach: GRPO stability and variants · 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
scale token loss by inverse process-set size (lambda-GRPO) (ablation)GRPO · toy tree n=1, r_neg=-1.0target path frequency: 0.0000 -> 0.7535Improved
scale token loss by inverse process-set size (ablation)GRPO · toy tree n=2, r_neg=-1.5target path frequency: 0.0000 -> 0.3820Improved
scale token loss by inverse process-set size (ablation)GRPO · toy tree n=1, r_neg=-0.5target path frequency: 0.3995 -> 0.5663Improved
scale token loss by inverse process-set size (ablation)GRPO · Qwen average accuracy, beta=0.0accuracy: 0.4837 -> 0.5576Improved
scale token loss by inverse process-set size (ablation)GRPO · Qwen average accuracy, beta=0.04accuracy: 0.5394 -> 0.5739Improved
scale token loss by inverse process-set size (ablation)GRPO · Llama average accuracy, beta=0.0accuracy: 0.0842 -> 0.1001Improved
scale token loss by inverse process-set size (ablation)GRPO · Llama average accuracy, beta=0.04accuracy: 0.0996 -> 0.0973Hurt
scale token loss by inverse process-set size (ablation)GRPO · Qwen MATH-500, beta=0.04accuracy: 0.8660 -> 0.8340Hurt
scale token loss by inverse process-set size (ablation)GRPO · Qwen AIME24, beta=0.04accuracy: 0.3000 -> 0.4000Improved
scale token loss by inverse process-set size (ablation)GRPO · Llama AMC23, beta=0.04accuracy: 0.1750 -> 0.0750Hurt
scale token loss by inverse process-set size (ablation)GRPO · validation accuracy during trainingvalidation accuracy: more than 10% increase in less than half the training stepsImproved
train Qwen with GRPO instead of leaving base model (beta=0.0)DeepSeek-R1-Distill-Qwen-1.5B · average accuracyaccuracy: 0.5175 -> 0.4837Hurt

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