AI Paper Atlas

Research map / Reinforcement learning / RL from feedback

Scaf-GRPO: Scaffolded Group Relative Policy Optimization for Enhancing LLM Reasoning

Xichen Zhang, Sitong Wu, Yinghao Zhu, Haoru Tan, Shaozuo Yu, Ziyi He, Jiaya Jia

ICLR 2026 · Reinforcement learning › RL from feedback · approach: GRPO stability and variants · 38 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
add scaffolded hints (Scaf-GRPO) to vanilla GRPOQwen2.5-Math-7B · Math benchmarks averagepass@1 (%): 45.2 -> 50.9Improved
add Scaf-GRPO to vanilla GRPOQwen2.5-Math-1.5B · Math benchmarks averagepass@1 (%): 37.6 -> 41.5Improved
add Scaf-GRPO to vanilla GRPOLlama-3.2-3B-Instruct · Math benchmarks averagepass@1 (%): 26.1 -> 28.8Improved
add Scaf-GRPO to vanilla GRPODeepSeek-R1-Distill-Qwen-1.5B · Math benchmarks averagepass@1 (%): 50.6 -> 53.6Improved
use progressive hierarchy K->P->S instead of solution-only hint (ablation)Scaf-GRPO · Math benchmarks averagepass@1 (%): 48.4 -> 50.9Improved
add the solution hint to the K->P hierarchy (ablation)Scaf-GRPO · Math benchmarks averagepass@1 (%): 48.0 -> 50.9Improved
add the knowledge hint to the P->S hierarchy (ablation)Scaf-GRPO · Math benchmarks averagepass@1 (%): 49.2 -> 50.9Improved
add the planning hint to the K->S hierarchy (ablation)Scaf-GRPO · Math benchmarks averagepass@1 (%): 48.6 -> 50.9Improved
deliver hints incrementally in chunks instead of full hint at once (ablation)Scaf-GRPO · Math benchmarks averagepass@1 (%): 47.7 -> 50.9Improved
add a guidance exemption period (no scaffolding from the very beginning vs exemption) (ablation)Scaf-GRPO · Math benchmarks averagepass@1 (%)Improved
vary guidance exemption duration from 10% to 40% (ablation)Scaf-GRPO · Math benchmarks averagepass@1 (%): plateau between 49.5% and 50.9%No clear effect
train on harder filtered data (ablation)Vanilla GRPO · Math benchmarks average (Qwen2.5-Math-7B)pass@1 (%): 43.9 -> 44.1No 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