Phi 3: Definition and Examples

Phi 3 is the third generation of Microsoft's Phi series, a family of Small Language Models (SLMs) that pushes the boundaries of what compact models can achieve. Launched in April 2024, this family includes several variants — Phi-3-mini (3.8 billion parameters), Phi-3-small (7 billion), and Phi-3-medium (14 billion) — each optimized for a different balance between performance and efficiency.

The particularity of Phi 3 lies in its training methodology. Rather than simply increasing model size, Microsoft focused on the quality of training data by using carefully filtered datasets and synthetic data generated by larger models. This approach, inspired by the principle 'textbooks are all you need', allows Phi 3 to achieve performance comparable to much larger models on many benchmarks.

Phi-3-mini, the flagship model of the family, can run directly on a smartphone or laptop thanks to its small size. It supports a context window of up to 128,000 tokens and particularly excels in logical reasoning, mathematics, and code generation. This ability to run locally makes it a preferred choice for applications requiring data privacy or offline operation.

Available as open source under the MIT license, Phi 3 integrates easily into existing pipelines via platforms like Hugging Face, Ollama, or Azure AI. Its performance-to-size ratio makes it particularly relevant for developers and companies looking to deploy generative AI solutions without the infrastructure costs associated with large models.