analiza aiqbench
This is an AI research paper on reinforcement learning for LLM reasoning, so it passes the topic gate. Its core claim is that rubric-based rewards work better when scored per reasoning step rather than collapsed into one response-level number, with reported gains of roughly 2.75 to 3.57 accuracy points on math benchmarks and reduced self-correction looping. These are benchmark results on Qwen3 models from a single preprint, so readers should treat them as promising but unverified at scale, and the step-level judge's reliability is the key open question.
Step-wise Rubric Rewards for LLM Reasoning
arXiv:2605.17291v2 Announce Type: replace Abstract: Reinforcement Learning with Verifiable Rewards (RLVR) improves language-model reasoning using final-answer correctness, but lacks supervision for intermediate steps. Rubric-based methods such as Rubrics as Rewards (RaR) provide finer-grained criteria, yet collapse rubric scores into a response-level scalar. This loses criterion-level structure, rewards flawed steps in correct responses, penalizes valid steps in incorrect responses, and can encourage unbounded self-correction. In 1,000 sampled problems, 18.2% of steps in answer-correct responses are wrong, while 49.9% of steps in answer-incorrect responses are correct. We propose Step-wise Rubrics as Rewards (SRaR), which (i) uses an LLM judge to assign rubric items to reasoning steps, (ii) normalizes step scores across rollouts to focus learning on quality differences, and (iii) combines step-wise and outcome rewards through a decoupled advantage estimator that preserves a stable outcome baseline. We also construct a 16K-problem rubric dataset by contrastively distilling criteria from verified correct and flawed reasoning paths. Across six math benchmarks, SRaR improves average accuracy over RaR by 3.57 and 2.75 points on Qwen3-8B-Non-Thinking and Qwen3-32B-Non-Thinking, respectively. On AIME 2025, it increases the Faithful Reasoning Rate from 34.5% to 46.7% and reduces self-correction looping from 48.1% to 26.5%. SRaR also achieves the best results on three of four code-generation benchmarks.
Abstract:Reinforcement Learning with Verifiable Rewards (RLVR) improves language-model reasoning using final-answer correctness, but lacks supervision for intermediate steps. Rubric-based methods such as Rubrics as Rewards (RaR) provide finer-grained criteria, yet collapse rubric scores into a response-level scalar. This loses criterion-level structure, rewards flawed steps in correct responses, penalizes valid steps in incorrect responses, and can encourage unbounded self-correction. In 1,000 sampled problems, 18.2% of steps in answer-correct responses are wrong, while 49.9% of steps in answer-incorrect responses are correct. We propose Step-wise Rubrics as Rewards (SRaR), which (i) uses an LLM judge to assign rubric items to reasoning steps, (ii) normalizes step scores across rollouts to focus learning on quality differences, and (iii) combines step-wise and outcome rewards through a decoupled advantage estimator that preserves a stable outcome baseline. We also construct a 16K-problem rubric dataset by contrastively distilling criteria from verified correct and flawed reasoning paths. Across six math benchmarks, SRaR improves average accuracy over RaR by 3.57 and 2.75 points on Qwen3-8B-Non-Thinking and Qwen3-32B-Non-Thinking, respectively. On AIME 2025, it increases the Faithful Reasoning Rate from 34.5% to 46.7% and reduces self-correction looping from 48.1% to 26.5%. SRaR also achieves the best results on three of four code-generation benchmarks.
Machine Learning (cs.LG)
arXiv:2605.17291 [cs.LG]
(or arXiv:2605.17291v2 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2605.17291 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 292 cuvinte
Subiecte conexe
Stepwise Reasoning Enhancement for LLMs via External Subgraph Generation
arXiv · 4 iun. 2026
Rethinking Stepwise Model Routing: A Cost-Efficient Table Reasoning Perspective
arXiv · 3 nov. 2026
Beyond Access: Guided LLM Scaffolding for Independent Learning in Undergraduate Statistics
arXiv · 21 iul. 2026
Reward-SQL: Boosting Text-to-SQL via Stepwise Execution-Aware Reasoning and Process-Supervised Rewards
arXiv · 15 iun. 2026