Microsoft on Tuesday released an upgraded version of its small language model, Phi-3, that can handle many tasks that were previously thought to require larger models. "Phi-3 is not just slightly cheaper, it's significantly cheaper; we're talking about a 10x cost difference compared to other models of similar capabilities," said Sébastien Bubeck, vice president of GenAI research at Microsoft.
The company said SLM is designed to perform simpler tasks, making it easier to use for companies with limited resources.
The company said Phi-3-mini will be immediately available on Azure, Microsoft's cloud service platform, a catalog of artificial intelligence models, Hugging Face, a platform for machine learning models, and Ollama, a framework for running models on local machines.
"The latest 'small language model' comes from Microsoft peru mobile database with Phi-3 Mini. This is a lightweight AI model that is part of its Small Language Model SLM family. It is the first of three SLMs that Microsoft plans to launch in the near future, the other two being Phi-3 Small and Phi-3 Medium. With a parameter capacity of 3.8 billion, Phi-3 Mini is designed to perform simpler tasks, making it more accessible and affordable for enterprises with limited resourcesompared to larger language models like GPT-4, Phi-3 Mini works on smaller datasets. It’s part of Microsoft’s broader plan to launch a series of SLMs tailored for simpler tasks, making them ideal for businesses with fewer resources. This approach promises to reduce costs and make models faster because they’re at the edge, and is expected to enable more enterprise and consumer use cases for generative AI. “Phi-3 enables developers to build generative AI applications at the edge with greater efficiency, lower latency, and lower cost. Microsoft has been developing a series of SLMs that offer many of the same capabilities as LLMs, but in a smaller size and trained on smaller amounts of data,” said Sebastien Bubeck, vice president of Microsoft GenAI. “We started experimenting with small language models a year ago to see how far we could go with just a billion parameters.”
Microsoft's competitors have their own small AI models, most of which are aimed at simpler tasks such as document summarization or coding assistance. Google's Gemma 2B and 7B are suitable for simple chatbots and language-related work.