2024-05-13 22:57:33
Microsoft has announced the launch of its smallest artificial intelligence (AI) model (SLM).
As Day.Az reports with reference to Lenta.ru, The Verge reports this.
The presented Phi-3 Mini is based on 3.8 billion parameters. It was trained on a smaller dataset compared to Large Language Models (LLMs) such as GPT-4. The model will be available to users of Azure, Hugging Face and Ollama services.
Company specialists said that Phi-3 works noticeably better than the previous version of the model. It is also very efficient: Phi-3 can produce the same answers as models with about 10 times more parameters. According to corporation representative Eric Boyd, the new model is not inferior in its capabilities to the GPT-3.5 type LLM, but has a “smaller form factor.”
Boyd also said that the developers taught Phi-3 using a “curriculum.” Experts were inspired by the way children gain knowledge through bedtime stories – books with simpler words and sentence structure. “There weren’t enough children’s books, so we took a list of over 3,000 words and asked the engineers to write us bedtime stories,” Boyd concluded.
In December 2023, Microsoft introduced the Phi-2 model, and the corporation also plans to release the Phi-3 Small and Phi-3 Medium models in the near future.