Tiny but mighty: The Phi-3 small language models with big potential

Microsoft researchers have introduced the Phi-3 family of small language models (SLMs), which offer significant AI capabilities in a more compact, accessible, and cost-effective format compared to large language models (LLMs). The new models, including Phi-3-mini with 3.8 billion parameters, outperform models of comparable and larger size in various benchmarks. Inspired by early childhood learning, the Phi-3 models leverage innovative training data techniques and emphasize high-quality content to achieve their performance. These SLMs are now available through multiple platforms and are particularly suited for applications requiring swift responses, offline capabilities, and enhanced privacy, making AI accessible to more users and scenarios, such as rural areas with limited network access. Future Phi-3 models promise to expand the range of options available according to quality and cost, further diversifying the potential applications of AI in various environments.

Opens on Microsoft