Sticker-TTS: Learn to Utilize Historical Experience with a Sticker-driven Test-Time Scaling Framework
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866908524105170944 |
|---|---|
| author | Chen, Jie Jiang, Jinhao Min, Yingqian Dong, Zican Wang, Shijie Zhao, Wayne Xin Wen, Ji-Rong |
| author_facet | Chen, Jie Jiang, Jinhao Min, Yingqian Dong, Zican Wang, Shijie Zhao, Wayne Xin Wen, Ji-Rong |
| contents | Large reasoning models (LRMs) have exhibited strong performance on complex reasoning tasks, with further gains achievable through increased computational budgets at inference. However, current test-time scaling methods predominantly rely on redundant sampling, ignoring the historical experience utilization, thereby limiting computational efficiency. To overcome this limitation, we propose Sticker-TTS, a novel test-time scaling framework that coordinates three collaborative LRMs to iteratively explore and refine solutions guided by historical attempts. At the core of our framework are distilled key conditions-termed stickers-which drive the extraction, refinement, and reuse of critical information across multiple rounds of reasoning. To further enhance the efficiency and performance of our framework, we introduce a two-stage optimization strategy that combines imitation learning with self-improvement, enabling progressive refinement. Extensive evaluations on three challenging mathematical reasoning benchmarks, including AIME-24, AIME-25, and OlymMATH, demonstrate that Sticker-TTS consistently surpasses strong baselines, including self-consistency and advanced reinforcement learning approaches, under comparable inference budgets. These results highlight the effectiveness of sticker-guided historical experience utilization. Our code and data are available at https://github.com/RUCAIBox/Sticker-TTS. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2509_05007 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Sticker-TTS: Learn to Utilize Historical Experience with a Sticker-driven Test-Time Scaling Framework Chen, Jie Jiang, Jinhao Min, Yingqian Dong, Zican Wang, Shijie Zhao, Wayne Xin Wen, Ji-Rong Artificial Intelligence Computation and Language I.2.7 Large reasoning models (LRMs) have exhibited strong performance on complex reasoning tasks, with further gains achievable through increased computational budgets at inference. However, current test-time scaling methods predominantly rely on redundant sampling, ignoring the historical experience utilization, thereby limiting computational efficiency. To overcome this limitation, we propose Sticker-TTS, a novel test-time scaling framework that coordinates three collaborative LRMs to iteratively explore and refine solutions guided by historical attempts. At the core of our framework are distilled key conditions-termed stickers-which drive the extraction, refinement, and reuse of critical information across multiple rounds of reasoning. To further enhance the efficiency and performance of our framework, we introduce a two-stage optimization strategy that combines imitation learning with self-improvement, enabling progressive refinement. Extensive evaluations on three challenging mathematical reasoning benchmarks, including AIME-24, AIME-25, and OlymMATH, demonstrate that Sticker-TTS consistently surpasses strong baselines, including self-consistency and advanced reinforcement learning approaches, under comparable inference budgets. These results highlight the effectiveness of sticker-guided historical experience utilization. Our code and data are available at https://github.com/RUCAIBox/Sticker-TTS. |
| title | Sticker-TTS: Learn to Utilize Historical Experience with a Sticker-driven Test-Time Scaling Framework |
| topic | Artificial Intelligence Computation and Language I.2.7 |
| url | https://arxiv.org/abs/2509.05007 |