DAST: Difficulty-Adaptive Slow-Thinking for Large Reasoning Models
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , , , , , , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2025
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866914246270386176 |
|---|---|
| author | Shen, Yi Zhang, Jian Huang, Jieyun Shi, Shuming Zhang, Wenjing Yan, Jiangze Wang, Ning Wang, Kai Liu, Zhaoxiang Lian, Shiguo |
| author_facet | Shen, Yi Zhang, Jian Huang, Jieyun Shi, Shuming Zhang, Wenjing Yan, Jiangze Wang, Ning Wang, Kai Liu, Zhaoxiang Lian, Shiguo |
| contents | Recent advancements in slow thinking reasoning models have shown exceptional performance in complex reasoning tasks. However, these models often exhibit overthinking (generating redundant reasoning steps for simple problems), leading to excessive computational resource usage. While current mitigation strategies uniformly reduce reasoning tokens, they risk degrading performance on challenging tasks that require extended reasoning. This paper introduces Difficulty-Adaptive Slow Thinking (DAST), a novel framework that enables models to autonomously adjust the length of Chain-of-Thought (CoT) based on problem difficulty. We first propose a Token Length Budget (TLB) metric to quantify difficulty, then leverage budget-aware reward shaping and budget preference optimization to implement DAST. DAST penalizes overlong responses for simple tasks while incentivizing sufficient reasoning for complex problems. Experiments on diverse datasets and model scales demonstrate that DAST effectively mitigates overthinking (reducing token usage by over 30\% on average) while preserving reasoning accuracy on complex problems. Our codes and models are available at https://github.com/AnonymousUser0520/AnonymousRepo01. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2503_04472 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | DAST: Difficulty-Adaptive Slow-Thinking for Large Reasoning Models Shen, Yi Zhang, Jian Huang, Jieyun Shi, Shuming Zhang, Wenjing Yan, Jiangze Wang, Ning Wang, Kai Liu, Zhaoxiang Lian, Shiguo Machine Learning Artificial Intelligence Recent advancements in slow thinking reasoning models have shown exceptional performance in complex reasoning tasks. However, these models often exhibit overthinking (generating redundant reasoning steps for simple problems), leading to excessive computational resource usage. While current mitigation strategies uniformly reduce reasoning tokens, they risk degrading performance on challenging tasks that require extended reasoning. This paper introduces Difficulty-Adaptive Slow Thinking (DAST), a novel framework that enables models to autonomously adjust the length of Chain-of-Thought (CoT) based on problem difficulty. We first propose a Token Length Budget (TLB) metric to quantify difficulty, then leverage budget-aware reward shaping and budget preference optimization to implement DAST. DAST penalizes overlong responses for simple tasks while incentivizing sufficient reasoning for complex problems. Experiments on diverse datasets and model scales demonstrate that DAST effectively mitigates overthinking (reducing token usage by over 30\% on average) while preserving reasoning accuracy on complex problems. Our codes and models are available at https://github.com/AnonymousUser0520/AnonymousRepo01. |
| title | DAST: Difficulty-Adaptive Slow-Thinking for Large Reasoning Models |
| topic | Machine Learning Artificial Intelligence |
| url | https://arxiv.org/abs/2503.04472 |