Microsoft on Tuesday unveiled Phi-3, its smallest language artificial intelligence (AI) model to date. Smaller AI models are significant because they can run on smartphones. The latest AI model is the successor to Phi-2, published in December 2023 and has a more extensive training database and settings. The enhanced parameters allow the AI model to understand and reply to more complicated questions than its predecessor. It is also comparable to models trained with over ten times the parameters utilised in Phi-3.
A preprint study describing the small language model (SLM) has been released on arXiv. However, because arXiv does not undertake peer reviews, the legitimacy of the assertions remains to be determined. AI enthusiasts can test the AI model using Azure and Ollama. Microsoft stated that the AI model is also available on the Nvidia NIM microservice, which has a standard API interface and is tailored for Nvidia GPUs. A Hugging Face catalogue for the Phi-3-mini has also been made, though the weights have yet to be released.
Phi-3: High Performance
In terms of performance, the AI model was trained using 3.3 trillion tokens, data units that include words, sentences, or subsections of words and are input into the system to train an AI model. It also has 3.8 billion parameters, demonstrating the level of complexity that the chatbot can understand. They are essentially neural connections in which each point represents knowledge about a specific topic and connects to various other points containing information relevant to the original point.
Phi-3 showcases exceptional performance metrics, surpassing models of comparable size and even outperforming some larger counterparts across a spectrum of benchmarks. Whether it's language comprehension, logical reasoning, coding proficiency, or mathematical capabilities, Phi-3 consistently demonstrates superior performance, making it a formidable contender in artificial intelligence.
Based on internal benchmarks, Microsoft states that the Chabot competes with models such as Mixtral 8x7B and GPT-3.5, which are significantly larger than the SML. The AI is designed to answer conversational requests in chat format. "We also provide some initial parameter-scaling results with 7B and 14B models trained for 4.8T tokens, called phi-3-small and phi-3-medium, both significantly more capable than phi-3-mini," the company adds.
Microsoft has made Phi-3 available as an open-source project.
In a strategic move to foster collaboration and innovation, Microsoft has made Phi-3 available as an open-source project. This means developers, researchers, and AI enthusiasts worldwide have unrestricted access to the model's architecture, source code, and documentation. By embracing an open-source approach, Microsoft empowers the global AI community to explore, experiment, and enhance Phi-3, driving forward advancements in natural language processing and AI technology as a whole.