Saved in:
| Main Authors: | , , , , , , , , , , , , , , , , , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2504.21318 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866916714510286848 |
|---|---|
| author | Abdin, Marah Agarwal, Sahaj Awadallah, Ahmed Balachandran, Vidhisha Behl, Harkirat Chen, Lingjiao de Rosa, Gustavo Gunasekar, Suriya Javaheripi, Mojan Joshi, Neel Kauffmann, Piero Lara, Yash Mendes, Caio César Teodoro Mitra, Arindam Nushi, Besmira Papailiopoulos, Dimitris Saarikivi, Olli Shah, Shital Shrivastava, Vaishnavi Vineet, Vibhav Wu, Yue Yousefi, Safoora Zheng, Guoqing |
| author_facet | Abdin, Marah Agarwal, Sahaj Awadallah, Ahmed Balachandran, Vidhisha Behl, Harkirat Chen, Lingjiao de Rosa, Gustavo Gunasekar, Suriya Javaheripi, Mojan Joshi, Neel Kauffmann, Piero Lara, Yash Mendes, Caio César Teodoro Mitra, Arindam Nushi, Besmira Papailiopoulos, Dimitris Saarikivi, Olli Shah, Shital Shrivastava, Vaishnavi Vineet, Vibhav Wu, Yue Yousefi, Safoora Zheng, Guoqing |
| contents | We introduce Phi-4-reasoning, a 14-billion parameter reasoning model that achieves strong performance on complex reasoning tasks. Trained via supervised fine-tuning of Phi-4 on carefully curated set of "teachable" prompts-selected for the right level of complexity and diversity-and reasoning demonstrations generated using o3-mini, Phi-4-reasoning generates detailed reasoning chains that effectively leverage inference-time compute. We further develop Phi-4-reasoning-plus, a variant enhanced through a short phase of outcome-based reinforcement learning that offers higher performance by generating longer reasoning traces. Across a wide range of reasoning tasks, both models outperform significantly larger open-weight models such as DeepSeek-R1-Distill-Llama-70B model and approach the performance levels of full DeepSeek-R1 model. Our comprehensive evaluations span benchmarks in math and scientific reasoning, coding, algorithmic problem solving, planning, and spatial understanding. Interestingly, we observe a non-trivial transfer of improvements to general-purpose benchmarks as well. In this report, we provide insights into our training data, our training methodologies, and our evaluations. We show that the benefit of careful data curation for supervised fine-tuning (SFT) extends to reasoning language models, and can be further amplified by reinforcement learning (RL). Finally, our evaluation points to opportunities for improving how we assess the performance and robustness of reasoning models. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2504_21318 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Phi-4-reasoning Technical Report Abdin, Marah Agarwal, Sahaj Awadallah, Ahmed Balachandran, Vidhisha Behl, Harkirat Chen, Lingjiao de Rosa, Gustavo Gunasekar, Suriya Javaheripi, Mojan Joshi, Neel Kauffmann, Piero Lara, Yash Mendes, Caio César Teodoro Mitra, Arindam Nushi, Besmira Papailiopoulos, Dimitris Saarikivi, Olli Shah, Shital Shrivastava, Vaishnavi Vineet, Vibhav Wu, Yue Yousefi, Safoora Zheng, Guoqing Artificial Intelligence Computation and Language We introduce Phi-4-reasoning, a 14-billion parameter reasoning model that achieves strong performance on complex reasoning tasks. Trained via supervised fine-tuning of Phi-4 on carefully curated set of "teachable" prompts-selected for the right level of complexity and diversity-and reasoning demonstrations generated using o3-mini, Phi-4-reasoning generates detailed reasoning chains that effectively leverage inference-time compute. We further develop Phi-4-reasoning-plus, a variant enhanced through a short phase of outcome-based reinforcement learning that offers higher performance by generating longer reasoning traces. Across a wide range of reasoning tasks, both models outperform significantly larger open-weight models such as DeepSeek-R1-Distill-Llama-70B model and approach the performance levels of full DeepSeek-R1 model. Our comprehensive evaluations span benchmarks in math and scientific reasoning, coding, algorithmic problem solving, planning, and spatial understanding. Interestingly, we observe a non-trivial transfer of improvements to general-purpose benchmarks as well. In this report, we provide insights into our training data, our training methodologies, and our evaluations. We show that the benefit of careful data curation for supervised fine-tuning (SFT) extends to reasoning language models, and can be further amplified by reinforcement learning (RL). Finally, our evaluation points to opportunities for improving how we assess the performance and robustness of reasoning models. |
| title | Phi-4-reasoning Technical Report |
| topic | Artificial Intelligence Computation and Language |
| url | https://arxiv.org/abs/2504.21318 |