Microsoft has launched Phi-3 Mini, a cost-effective and powerful small language model with capabilities in content creation, social media post generation, and more, all while requiring less data compared to its larger counterparts.
Phi-3 Mini is the first of three small AI models planned for release by Microsoft. Clocking in at 3.8 billion parameters, it’s smaller than models like GPT-4, but comes with a punch in performance.
An advantage of the Phi-3 Mini includes being trained on a smaller dataset, making it more efficient to run. It has lower operational costs for businesses and performs smoothly on personal devices like laptops and smartphones.
Smaller models like Phi-3 Mini are more accessible to companies with limited resources. Because they require less data and computing power, they are a more cost-effective option for many businesses.
Microsoft claims it can outperform models twice its size on various benchmarks that evaluate language, coding, and math capabilities. The company even notes that Phi-3 Mini’s capabilities approach those of models ten times larger.
Compared to others like Phi-1 which focused on coding, and Phi-2 which began to develop reasoning skills, Phi-3 Mini refines these abilities, bringing increased competence in both coding and reasoning tasks.
An interesting aspect of Phi-3 Mini’s development involves its training method. Inspired by how children learn, developers exposed the model to “bedtime stories” and simpler texts alongside materials that tackled broader subjects. Essentially, Phi-3 Mini learned foundational concepts before progressing to more complex information.
While smaller models like Phi-3 Mini may not possess the large knowledge base of models trained on the entirety of the internet (like GPT-4), they offer a distinct advantage for many businesses. Companies often have internal datasets that are smaller in scope.
Phi-3 Mini’s focus and efficiency make it ideal for working with such datasets, allowing businesses to leverage AI power for tasks like summarizing documents or extracting insights from market research reports.
Phi-3 Mini is available immediately on several platforms including Azure’s AI model catalog — Microsoft’s cloud service platform, Hugging Face — a machine learning model platform, Ollama — a framework for running models on local machines and NVIDIA NIM, where it was optimized for peak performance on NVIDIA GPUs, with a standard API interface.
This launch is the beginning of a series of small AI models from Microsoft. Phi-3 Mini is followed by Phi-3 Small (7 billion parameters) and Phi-3 Medium (14 billion parameters), offering businesses a range of options to suit their specific needs.
Microsoft’s latest development follows the company’s $1.5 billion investment in UAE-based AI firm G42 and its partnership with French startup Mistral AI to bring their models to the Azure cloud computing platform.
