Microsoft launches Phi-3, A Lightweight Language Model but Super Capable

Microsoft announced a small artificial intelligence language model that is capable of doing tasks like writing social media posts and creating content using less amounts of data.

The tech giant said in a statement that their new lightweight model called Phi-3-mini has the ability to outdo bigger models ten times its size when tested for a range of benchmarks like coding, math, and language capabilities. 

Buy physical gold and silver online
MSFT share price. Source: NASDAQ via search.

Phi-3 is a suitable solution for smaller businesses

Microsoft said that these types of smaller models make it easier for smaller firms with limited resources to perform simpler tasks, as they are designed with this approach in mind.

Phi-3 could be used by a business to extract relevant information and business trends from market research reports and to fetch and summarize the main points of a long document, explained Microsoft.

Phi-3 will be available instantly on the AI model catalog of Azur, which is Microsoft’s cloud service platform, on Hugging Face, which is a machine learning model testing and deployment platform, and also on Ollama, which is a platform that provides frameworks for running models.

More Phi-3 models will come with more capabilities

Microsoft Azure’s flowchart for candidate compute services. Source: Microsoft.

Phi-3 is capable of 3.8 billion parameters and is trained on quite less data if compared to large language models like Gemini or GPT-4. Microsoft released Phi-2 at the end of the previous year. Experts say that it performed just as good as some of the larger models. Phi-3 is expected to perform even much better than its predecessor. 

Microsoft plans to introduce two more versions of the model, as it is the first in the lineup, a Phi-3 Small and a Phi-3 Medium, with different capabilities of parameters from seven to 14 billion.

The lightweight model cannot generate results like those of cloud-based LLMs, but it can compete with other smaller models like Mistral and llama-3 in writing codes, solving mathematical equations, and academic papers. 

Another added benefit of the Phi-3 is that it can run without an internet connection and that on smaller devices like a cell phone. Phi-3 is a cost-effective solution for smaller companies that cannot afford cloud-based large language models to get into AI.

The original news from Reuters can be seen here, and Microsoft’s research paper is here.

About the author

Why invest in physical gold and silver?
文 » A