AI Paper Atlas

Research map / Reinforcement learning / RL from feedback

Unlocking Token Rewards via Training-Free Reward Attribution

Sitong Wu, Haoru Tan, Bin Xia, Xichen Zhang, Jingyao Li, Shaofeng Zhang, Xiaojuan Qi, Bei Yu, Jiaya Jia

CVPR 2026 · Reinforcement learning › RL from feedback · approach: Scaling RL with verifiable rewards · 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
add P2T token-level reward to GRPOQwen2.5-Math-7B · 7 math benchmarks avgPass@1: 48.0 -> 53.8Improved
add P2T token-level reward to GRPOLLaMA3.2-3B-Instruct · 7 math benchmarks avgPass@1: 26.4 -> 35.8Improved
add P2T token-level reward to GRPODeepSeek-R1-Distill-Qwen-1.5B · 7 math benchmarks avgPass@1: 50.4 -> 57.2Improved
add P2T token-level reward to GRPOQwen3-1.7B (non-thinking mode) · 7 math benchmarks avgPass@1: 47.8 -> 53.0Improved
add P2T token reward versus PRIME and SPRO token rewardsEurus-2-7B-SFT · Eurus-2 math and code avgPass@1: 38.4 -> 40.7 (SPRO), 36.0 -> 40.7 (PRIME)Improved
add token-level redistribution of step reward within each step (vs step reward only) (ablation)GRPO + PRM reward, Qwen2.5-Math-7B · 7 math benchmarks avgPass@1: 50.6 -> 53.8Improved
add step-level process reward to outcome reward (ablation)GRPO, Qwen2.5-Math-7B · 7 math benchmarks avgPass@1: 48.0 -> 50.6Improved
vary redistribution strength eta from 0.25 to 1.0 (ablation)P2T · 7 math benchmarks avgPass@1: 53.5 / 53.8 / 53.4 / 53.2No clear effect
use signed redistribution instead of magnitude-based (ablation)P2T · 7 math benchmarks avgPass@1: 53.8 -> 52.6Hurt
use stronger PRM (ReasonFlux-PRM-7B vs Skywork PRM-7B) (ablation)P2T text-only · 7 math benchmarks avgPass@1: 50.0 -> 53.8Improved
use VisualPRM instead of MM-PRM (ablation)P2T Qwen2.5-VL-7B-Instruct · 5 multimodal benchmarks avgPass@1: 55.3 -> 57.5Improved
use pad token as null token instead of EOS token (ablation)P2T · 7 math benchmarks avgPass@1: 50.3 -> 53.8Improved

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