Sarah CarterMicrosoft’s Tiny but Mighty AI Model That’s Set to Disrupt the Industry!
Microsoft has recently introduced Phi-3 Mini, the first of its kind from among the three AI models designed for customers around the world. Microsoft has recently introduced Phi-3 Mini, the first of its kind from among the three AI models designed for customers around the world. Phi-3 is an innovative model that has achieved a significant milestone in Microsoft’s efforts to democratize AI technology, making it an accessible and affordable option which is feature-rich & user-friendly.
For now, it’s set in stone to revolutionize the way we interact with AI.
What is the Phi-3 Mini?
Phi-3 Mini is a small language model (SLM) introduced by Microsoft, which is part of the Phi-3 family of open models. It has 3.8 billion parameters, making it smaller and more cost-effective compared to large language models (LLMs) like GPT-4.
Key Features and Capabilities
- Phi-3 Mini is designed for simpler tasks and can perform well on personal devices like phones and laptops.
- It has a 4,000-token context window, with a 128K-token version also available.
- Despite its smaller size, Phi-3 Mini can provide responses close to those of larger models, making it a more accessible option.
- It is currently available on Azure, Hugging Face, and Ollama, with plans to release Phi-3 Small (7 billion parameters) and Phi-3 Medium (14 billion parameters) in the future.
Microsoft plans to release two additional versions: Phi-3 Small (based on 7 billion parameters) and Phi-3 Medium (based on 14 billion parameters). The increased parameters will enhance the AI model’s understanding and result production capabilities.
Phi-2, launched in December, demonstrated impressive performance, outperforming Llama 2 in tests. Microsoft claims Phi-3 surpasses its predecessor, delivering approximately 10 times better results. Despite being labeled “tiny,” Phi-3 Mini packs a punch in producing relevant results.
According to Eric Boyd, Corporate Vice President of Microsoft Azure AI Platform, Phi-3 Mini can hold its own against larger language models like GPT 3.5, albeit on a smaller scale.
LLMs offer exciting opportunities, but their size requires significant computing resources. Microsoft’s SLMs, like Phi-3 Mini, offer similar capabilities at a smaller size, making them a more accessible option.
Early this year, The Information issued a report that talks about how Microsoft is putting together a team focused on building light-weight AI models. Today, we see Phi hitting the market, and that’s not everything that Microsoft has released. It has also launched Orca-Math, a model particularly focused on solving math problems for struggling students.
Many of Microsoft’s competitors have come forth with their own small AI models to beat the competition. Some of them are capable of solving simpler tasks while others are designed to take care of more complex ones such as document summarization or in fact, completely coding an app or solution.
For instance, Google’s Gemma 2B and 7B have better use for simple tasks such as the ones which are chatbot related. Similarly, Anthropic’s Claude 3 Haiku is designed to explore research, provide a summary and add graphs within them. Also, the recently released Llama 3 8B from Meta also contributes sequentially.
For now, all that we know is that Phi-3 has been trained based on a curriculum.
The Phi-3 family was simply built iterations from the previously released models. For instance, Phi-1 was merely focused on coding; whereas Phi-2 went a step ahead and learned how to reason.
With the Phi-3 release, we get a more robust solution that delivers optimized coding experience & offers better reasoning.
In the future, we can expect Phi-4 to have much better reasoning capabilities and produce better results overall.
Want to Know More About AI? Dive deeper into the world of artificial intelligence on Techvizer.
Explore our AI section for the latest news, trends, and insights.
