Saved in:
| Main Authors: | , , , , , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2505.24863 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866910976481165312 |
|---|---|
| author | Zhang, Junyu Dong, Runpei Wang, Han Ning, Xuying Geng, Haoran Li, Peihao He, Xialin Bai, Yutong Malik, Jitendra Gupta, Saurabh Zhang, Huan |
| author_facet | Zhang, Junyu Dong, Runpei Wang, Han Ning, Xuying Geng, Haoran Li, Peihao He, Xialin Bai, Yutong Malik, Jitendra Gupta, Saurabh Zhang, Huan |
| contents | This paper presents AlphaOne ($α$1), a universal framework for modulating reasoning progress in large reasoning models (LRMs) at test time. $α$1 first introduces $α$ moment, which represents the scaled thinking phase with a universal parameter $α$. Within this scaled pre-$α$ moment phase, it dynamically schedules slow thinking transitions by modeling the insertion of reasoning transition tokens as a Bernoulli stochastic process. After the $α$ moment, $α$1 deterministically terminates slow thinking with the end-of-thinking token, thereby fostering fast reasoning and efficient answer generation. This approach unifies and generalizes existing monotonic scaling methods by enabling flexible and dense slow-to-fast reasoning modulation. Extensive empirical studies on various challenging benchmarks across mathematical, coding, and scientific domains demonstrate $α$1's superior reasoning capability and efficiency. Project page: https://alphaone-project.github.io/ |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2505_24863 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | AlphaOne: Reasoning Models Thinking Slow and Fast at Test Time Zhang, Junyu Dong, Runpei Wang, Han Ning, Xuying Geng, Haoran Li, Peihao He, Xialin Bai, Yutong Malik, Jitendra Gupta, Saurabh Zhang, Huan Computation and Language This paper presents AlphaOne ($α$1), a universal framework for modulating reasoning progress in large reasoning models (LRMs) at test time. $α$1 first introduces $α$ moment, which represents the scaled thinking phase with a universal parameter $α$. Within this scaled pre-$α$ moment phase, it dynamically schedules slow thinking transitions by modeling the insertion of reasoning transition tokens as a Bernoulli stochastic process. After the $α$ moment, $α$1 deterministically terminates slow thinking with the end-of-thinking token, thereby fostering fast reasoning and efficient answer generation. This approach unifies and generalizes existing monotonic scaling methods by enabling flexible and dense slow-to-fast reasoning modulation. Extensive empirical studies on various challenging benchmarks across mathematical, coding, and scientific domains demonstrate $α$1's superior reasoning capability and efficiency. Project page: https://alphaone-project.github.io/ |
| title | AlphaOne: Reasoning Models Thinking Slow and Fast at Test Time |
| topic | Computation and Language |
| url | https://arxiv.org/abs/2505.24863 |