AI Paper Atlas

Research map / Reinforcement learning / RL from feedback

GTPO and GRPO-S: Token and Sequence-Level Reward Shaping with Policy Entropy

Hongze Tan, Zihan Wang, Jianfei Pan, Jinghao Lin, Hao Wang, Yifan Wu, Tao Chen, Zhihang Zheng, Zhihao Tang, Haihua Yang

ICML 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
use GTPO entropy-weighted token rewards instead of DAPOQwen2.5-7B · AIME 2024Pass@32: 35.51 -> 64.89Improved
use GTPO instead of GRPOQwen2.5-7B · AIME 2024Pass@32: 46.01 -> 64.89Improved
use GTPO instead of DAPOQwen2.5-32B · AIME 2024Pass@32: 59.02 -> 68.91Improved
use GTPO instead of DAPO w/ Forking TokensQwen2.5-7B · AIME 2024Pass@32: 54.16 -> 64.89Improved
use GRPO-S instead of DAPOQwen2.5-7B · AIME 2025Mean@32: 13.33 -> 18.33Improved
use GTPO instead of DAPOQwen2.5-32B · AIME 2025Pass@32: 26.67 -> 36.67Improved
use GTPO instead of DAPOQwen2.5-7B · MATH 500Mean@32: 78.85 -> 80.16Improved
use GRPO-S instead of DAPOQwen2.5-32B · MATH 500Mean@32: 84.45 -> 85.79Improved
use GTPO on the smaller 7B model rather than 32B (gain in Pass@32 over baseline)GTPO · AIME 2024Pass@32: +29.4 points on the 7B model, compared to +9.9 points on the 32B modelImproved
increase entropy bonus weight beta_2 from 0.1 to 0.2 (ablation)GRPO-S · AIME 2024Mean@32 and Pass@32: 0.1 -> 0.2: clear decline in overall performanceHurt
vary reward shaping hyperparameters across tested configurations (ablation)GTPO and GRPO-S · AIME 2024Mean@32 and Pass@32Improved

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