Phi-4-reasoning Technical Report
Marah Abdin, Sahaj Agarwal, Ahmed Awadallah, Vidhisha Balachandran, Harkirat Behl, Lingjiao Chen, Gustavo de Rosa, Suriya Gunasekar, Mojan Javaheripi, Neel Joshi, Piero Kauffmann, Yash Lara, Caio César Teodoro Mendes, Arindam Mitra, Besmira Nushi, Dimitris Papailiopoulos, Olli Saarikivi, Shital Shah, Vaishnavi Shrivastava, Vibhav Vineet, Yue Wu, Safoora Yousefi
2025-05-01
Summary
This paper talks about Phi-4-reasoning, a new AI model with 14 billion parameters that has been specially trained to handle tough reasoning problems better than even bigger models.
What's the problem?
Many large AI models are not very good at solving complex reasoning tasks, even though they have lots of data and computing power, which means they can still make mistakes on tricky questions.
What's the solution?
The researchers improved Phi-4-reasoning by using supervised fine-tuning and reinforcement learning, which are special training methods that help the model learn to reason more accurately and reliably.
Why it matters?
This matters because it shows that with the right training, smaller models can beat much bigger ones at difficult tasks, making advanced AI more efficient and accessible for things like homework help, research, and problem-solving.
Abstract
Phi-4-reasoning, a 14-billion parameter model enhanced with supervised fine-tuning and reinforcement learning, outperforms larger models on complex reasoning tasks across various benchmarks.