Microsoft announced the discharge of Phi-3 today a strong 3 billion parameter language model that gives advanced reasoning capabilities just like much larger models at a significantly lower cost. Developed by Microsoft Research, the brand new model might be available on the corporate's Azure AI platform and can enable firms to leverage state-of-the-art natural language processing and reasoning for various applications.
“What matters is that we’re in a position to have a really small model with capabilities – by way of benchmarks, by way of experience fiddling with the model – that may compete with much, much larger models, including approaching something like a GPT -3.5 level,” SĂ©bastien Bubeck, vice chairman of generative AI at Microsoft, told VentureBeat. “That’s really what that is about. It's not necessarily the type of progress we expected. I don’t think anyone knew what size you would want to get features near something like GPT-3.5.”
Phi-3 marks the most recent success in Microsoft's efforts to explore the boundaries of compact language models. Starting with the coding oriented Phi-1 a 12 months ago and is making progress Phi-1.5 And Phi-2The Phi series has demonstrated impressive performance in coding, common sense, and general natural language benchmarks with models as small as 1 to 2 billion parameters.
Enables cost-effective AI for businesses
“When customers saw what was possible, everyone ran and said, 'Okay, now I actually have to do something interesting with this,'” Eric Boyd, corporate vice chairman of the Azure AI Platform, told VentureBeat. “On Azure, we help these customers construct the generative AI applications they need…We will at all times have probably the most powerful models available that really push the boundaries and show the boundaries of what is feasible. But along the way in which we will even have the perfect model in every price segment.”
With Phi-3, Microsoft has developed a universal 3 billion parameter model that has comprehensive capabilities near industry-leading models akin to OpenAI GPT 3.5, but at a much lower cost and with the pliability to run on commodity hardware and even smartphones. This breakthrough in parameter efficiency enables transformative AI use cases for businesses that were previously unaffordable.
Responsible AI on the forefront
Microsoft developed Phi-3 with its Responsible AI principles in mind from the beginning. The model training data was reviewed for toxicity and bias, and extra safety measures were taken prior to publication. This allows firms, particularly those in regulated industries, to securely leverage Phi-3's capabilities.
From a technical perspective, Phi-3 runs on the ONNX runtime Optimized for NVIDIA GPUs and will be deployed across multiple GPUs or machines to optimize throughput. The model's architecture leverages efficient attention mechanisms and optimized numerical precision to realize high performance with a comparatively small variety of parameters.
Empowering businesses with advanced natural language AI
“The beauty is that now that you will have this fundamental layer in a small model, you may herald your data and refine this general model and get amazing performance on narrow verticals,” Bubeck explained. “Even if one chooses a narrow domain, the overall intelligence must even be good in that vertical domain.”
Microsoft's launch of Phi-3 and its planned integration with the Azure AI platform represent a big step forward in making the capabilities of enormous language models accessible and cost-effective for firms of all sizes. As more firms look to operationalize AI and unlock the worth of their unstructured data, purpose-built models like Phi-3 might be critical to achieving this vision.