Microsoft Unveils Phi-4: A Small Language Model Excelling in Complex Reasoning
2024-12-15
Microsoft has introduced Phi-4, a new 14-billion parameter small language model (SLM) that outperforms larger models in complex reasoning tasks, particularly in mathematics, surpassing even Gemini Pro 1.5 on math competition problems. This achievement is attributed to high-quality synthetic and organic datasets and post-training innovations. Currently available on Azure AI Foundry under an MSRLA, Phi-4 will launch on Hugging Face next week. Microsoft emphasizes its commitment to responsible AI development, integrating robust safety features into Phi-4's design and deployment.