Less is More Tokens: Efficient Math Reasoning via Difficulty-Aware Chain-of-Thought Distillation
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2025
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866914024347664384 |
|---|---|
| author | Waheed, Abdul Mitra, Chancharik Wang, Laurie Z. Ramanan, Deva Raj, Bhiksha |
| author_facet | Waheed, Abdul Mitra, Chancharik Wang, Laurie Z. Ramanan, Deva Raj, Bhiksha |
| contents | Chain-of-thought reasoning, while powerful, can produce unnecessarily verbose output for simpler problems. We present a framework for difficulty-aware reasoning that teaches models to dynamically adjust reasoning depth based on problem complexity. Remarkably, we show that models can be endowed with such dynamic inference pathways without any architectural modifications; we simply post-train on data that is carefully curated to include chain-of-thought traces that are proportional in length to problem difficulty. Our analysis reveals that post-training via supervised fine-tuning (SFT) primarily captures patterns like reasoning length and format, while direct preference optimization (DPO) preserves reasoning accuracy, with their combination reducing length and maintaining or improving performance. Both quantitative metrics and qualitative assessments confirm that models can learn to "think proportionally", reasoning minimally on simple problems while maintaining depth for complex ones. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2509_05226 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Less is More Tokens: Efficient Math Reasoning via Difficulty-Aware Chain-of-Thought Distillation Waheed, Abdul Mitra, Chancharik Wang, Laurie Z. Ramanan, Deva Raj, Bhiksha Computation and Language Chain-of-thought reasoning, while powerful, can produce unnecessarily verbose output for simpler problems. We present a framework for difficulty-aware reasoning that teaches models to dynamically adjust reasoning depth based on problem complexity. Remarkably, we show that models can be endowed with such dynamic inference pathways without any architectural modifications; we simply post-train on data that is carefully curated to include chain-of-thought traces that are proportional in length to problem difficulty. Our analysis reveals that post-training via supervised fine-tuning (SFT) primarily captures patterns like reasoning length and format, while direct preference optimization (DPO) preserves reasoning accuracy, with their combination reducing length and maintaining or improving performance. Both quantitative metrics and qualitative assessments confirm that models can learn to "think proportionally", reasoning minimally on simple problems while maintaining depth for complex ones. |
| title | Less is More Tokens: Efficient Math Reasoning via Difficulty-Aware Chain-of-Thought Distillation |
| topic | Computation and Language |
| url | https://arxiv.org/abs/2509.05226 |