Microsoft Phi-3 Launched as Company’s Smallest Open-Source AI Model to Date

Microsoft on Tuesday released Phi-3, its smallest artificial intelligence (AI) language model to date. Smaller AI models are important, because they have the potential to work with smartphones. The latest AI model is the successor to Phi-2, which was launched in December 2023, and includes a higher training database and larger parameters. The increased parameters help the AI ​​model understand and answer more complex questions compared to its predecessor. It is also said to be on par with models trained in more than 10 times the number of parameters used by Phi-3.

A preprint paper detailing the Small Language Model (SLM) has been published on arXiv. However, since arXiv does not conduct peer reviews, the validity of the claims has not yet been determined. AI enthusiasts can test the AI ​​model using Azure and Ollama. Microsoft said the AI ​​model is also available in Nvidia's NIM microservice with a standard API interface and has been optimized for Nvidia GPUs. A Hugging Face catalog has also been created for the Phi-3-mini, but the weights have not yet been released.

In terms of performance, the AI ​​model was trained on 3.3 trillion tokens—units of data that include words, phrases, or subsections of words that are fed into the system to train an AI model. It also contains 3.8 billion parameters, highlighting the level of complexity that the chatbot can understand. They are essentially neural connections where each point is knowledge about a certain topic and connects to several other points that contain contextual information to the original point.

Microsoft claims, based on internal benchmarking, that the chabot rivals the likes of the Mixtral 8x7B and GPT-3.5, which are much larger than the SML. The AI ​​is aligned with the chat format, meaning it can respond to conversational queries. “We also provide some initial parameter scaling results with 7B and 14B models trained for 4.8T tiles, called phi-3-small and phi-3-medium, both significantly more capable than phi-3-mini,” he said. say the tech giant. he says

Reuters reports that the AI ​​model, designed to perform simpler tasks, is also hosted on Microsoft Azure and Ollama. The company has not yet shared details about Phi-3-mini's open source license. In particular, the Apache 2.0 license, which Grok AI recently issued, allows both academic and commercial use.