Leveraging Reasoning Model Answers to Enhance Non-Reasoning Model Capability
Fuente:
arXiv
Guardado en:
| Autores principales: | , , , , , , , |
|---|---|
| Formato: | Preprint |
| Publicado: |
2025
|
| Materias: | |
| Acceso en línea: | |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| _version_ | 1866908317262020608 |
|---|---|
| author | Wang, Haotian Zhao, Han Chen, Shuaiting Tian, Xiaoyu Zhao, Sitong Ji, Yunjie Peng, Yiping Li, Xiangang |
| author_facet | Wang, Haotian Zhao, Han Chen, Shuaiting Tian, Xiaoyu Zhao, Sitong Ji, Yunjie Peng, Yiping Li, Xiangang |
| contents | Recent advancements in large language models (LLMs), such as DeepSeek-R1 and OpenAI-o1, have demonstrated the significant effectiveness of test-time scaling, achieving substantial performance gains across various benchmarks. These advanced models utilize deliberate "thinking" steps to systematically enhance answer quality. In this paper, we propose leveraging these high-quality outputs generated by reasoning-intensive models to improve less computationally demanding, non-reasoning models. We explore and compare methodologies for utilizing the answers produced by reasoning models to train and improve non-reasoning models. Through straightforward Supervised Fine-Tuning (SFT) experiments on established benchmarks, we demonstrate consistent improvements across various benchmarks, underscoring the potential of this approach for advancing the ability of models to answer questions directly. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2504_09639 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Leveraging Reasoning Model Answers to Enhance Non-Reasoning Model Capability Wang, Haotian Zhao, Han Chen, Shuaiting Tian, Xiaoyu Zhao, Sitong Ji, Yunjie Peng, Yiping Li, Xiangang Computation and Language Recent advancements in large language models (LLMs), such as DeepSeek-R1 and OpenAI-o1, have demonstrated the significant effectiveness of test-time scaling, achieving substantial performance gains across various benchmarks. These advanced models utilize deliberate "thinking" steps to systematically enhance answer quality. In this paper, we propose leveraging these high-quality outputs generated by reasoning-intensive models to improve less computationally demanding, non-reasoning models. We explore and compare methodologies for utilizing the answers produced by reasoning models to train and improve non-reasoning models. Through straightforward Supervised Fine-Tuning (SFT) experiments on established benchmarks, we demonstrate consistent improvements across various benchmarks, underscoring the potential of this approach for advancing the ability of models to answer questions directly. |
| title | Leveraging Reasoning Model Answers to Enhance Non-Reasoning Model Capability |
| topic | Computation and Language |
| url | https://arxiv.org/abs/2504.09639 |