Microsoft Introduces their Smallest AI Model to Date
Microsoft has recently introduced Phi-3 Mini, a lightweight artificial intelligence model that packs a punch despite its small size. Smaller AI models are easier and cheaper to both create and run. The decreased size means they perform well on personal devices, making them easier to adapt and launch. However, there is potential for corporate use too. An organizations internal dataset tends to be smaller, so using a smaller AI model is a good fit for internal use. It is easier, faster to install, and more affordable.
Here are some key features of Phi-3 Mini:
- Model Parameters: Phi-3 Mini is trained on 3.8 billion parameters
- Performance: It achieves performance rivaling models ten times its size
- Compact Design: Remarkably, Phi-3 Mini is compact enough to run on smartphones and other local devices
- Innovative Training Data: You do not need architectural changes to train the data
- Uses Cases: Pho-3 Mini performs well in tasks ranging from math to programming to academic tests
- Availability: Currently available on Azure, Hugging Face, and Ollama platforms
Microsoft has plans to release two more small models; Phi-3 Small measures seven billion parameters and Phi-3 Medium measures fourteen billion parameters. Stay tuned to the blog for more details.
Read More: