Microsoft has recently unveiled its smallest language artificial intelligence (AI) model, Phi-3. This new AI model is a successor to Phi-2 and boasts higher training database and larger parameters. The release of Phi-3 is a significant step in the field of AI as it opens up possibilities for running AI models on smartphones.
The Phi-3 AI model has been trained on 3.3 trillion tokens and contains 3.8 billion parameters. These parameters enable the AI model to understand and respond to more complex questions, making it on par with models trained on much larger parameter numbers. Microsoft claims that the Phi-3 mini can rival models like Mixtral 8x7B and GPT-3.5 in terms of performance.
The AI model is specifically designed for chat formats, allowing it to respond to conversational queries effectively. Microsoft has also introduced Phi-3-small and Phi-3-medium models, which are trained on significantly more tokens and exhibit enhanced capabilities compared to Phi-3-mini. Furthermore, the Phi-3 model is hosted on Microsoft Azure and Ollama, providing easy access to AI enthusiasts.
AI enthusiasts can test out the Phi-3 model through Azure and Ollama platforms. A Hugging Face catalog for Phi-3-mini has been created, but the weights are yet to be released. While a pre-print paper detailing the small language model has been published on arXiv, it is important to note that arXiv does not conduct peer reviews, raising questions about the validity of the claims.
Microsoft’s Phi-3-mini AI model represents a significant advancement in the field of artificial intelligence. With its smaller size and improved performance, the Phi-3 model opens up possibilities for running AI models on smartphones and responding to more complex queries. While the model is yet to release its open source license details, it is poised to make a substantial impact in the AI industry.
Leave a Reply