Gespeichert in:
| Hauptverfasser: | , , , , , , , , , , , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2026
|
| Schlagworte: | |
| Online-Zugang: | https://arxiv.org/abs/2602.11570 |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866914324466892800 |
|---|---|
| author | Wang, Xiangfeng Guo, Hangyu Lai, Yanlin Huang, Mitt Zhao, Liang Yao, Chengyuan Zhang, Yinmin Han, Qi Ren, Xiaoxiao Yuan, Chun Xu, Tong Ge, Zheng Zhang, Xiangyu Jiang, Daxin |
| author_facet | Wang, Xiangfeng Guo, Hangyu Lai, Yanlin Huang, Mitt Zhao, Liang Yao, Chengyuan Zhang, Yinmin Han, Qi Ren, Xiaoxiao Yuan, Chun Xu, Tong Ge, Zheng Zhang, Xiangyu Jiang, Daxin |
| contents | While model-based verifiers are essential for scaling Reinforcement Learning with Verifiable Rewards (RLVR), current outcome-centric verification paradigms primarily focus on the consistency between the final result and the ground truth, often neglecting potential errors in the derivation process. This leads to assigning positive rewards to correct answers produced from incorrect derivations. To bridge this gap, we introduce PRIME, a benchmark for evaluating verifiers on Process-Outcome Alignment verification in Mathematics and Engineering. Curated from a comprehensive collection of college-level STEM problems, PRIME comprises 2,530 high-difficulty samples through a consistency-based filtering pipeline. Through extensive evaluation, we find that current verifiers frequently fail to detect derivation flaws. Furthermore, we propose a process-aware RLVR training paradigm utilizing verifiers selected via PRIME. This approach substantially outperforms the outcome-only verification baseline, achieving absolute performance gains of 8.29%, 9.12%, and 7.31% on AIME24, AIME25, and Beyond-AIME, respectively, for the Qwen3-14B-Base model. Finally, we demonstrate a strong linear correlation ($R^2 > 0.92$) between verifier accuracy on PRIME and RLVR training effectiveness, validating PRIME as a reliable predictor for verifier selection. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2602_11570 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | PRIME: A Process-Outcome Alignment Benchmark for Verifiable Reasoning in Mathematics and Engineering Wang, Xiangfeng Guo, Hangyu Lai, Yanlin Huang, Mitt Zhao, Liang Yao, Chengyuan Zhang, Yinmin Han, Qi Ren, Xiaoxiao Yuan, Chun Xu, Tong Ge, Zheng Zhang, Xiangyu Jiang, Daxin Computation and Language While model-based verifiers are essential for scaling Reinforcement Learning with Verifiable Rewards (RLVR), current outcome-centric verification paradigms primarily focus on the consistency between the final result and the ground truth, often neglecting potential errors in the derivation process. This leads to assigning positive rewards to correct answers produced from incorrect derivations. To bridge this gap, we introduce PRIME, a benchmark for evaluating verifiers on Process-Outcome Alignment verification in Mathematics and Engineering. Curated from a comprehensive collection of college-level STEM problems, PRIME comprises 2,530 high-difficulty samples through a consistency-based filtering pipeline. Through extensive evaluation, we find that current verifiers frequently fail to detect derivation flaws. Furthermore, we propose a process-aware RLVR training paradigm utilizing verifiers selected via PRIME. This approach substantially outperforms the outcome-only verification baseline, achieving absolute performance gains of 8.29%, 9.12%, and 7.31% on AIME24, AIME25, and Beyond-AIME, respectively, for the Qwen3-14B-Base model. Finally, we demonstrate a strong linear correlation ($R^2 > 0.92$) between verifier accuracy on PRIME and RLVR training effectiveness, validating PRIME as a reliable predictor for verifier selection. |
| title | PRIME: A Process-Outcome Alignment Benchmark for Verifiable Reasoning in Mathematics and Engineering |
| topic | Computation and Language |
| url | https://arxiv.org/abs/2602.11570 |