Microsoft’s new Phi-4 AI models pack big performance in small packages
1 min read
Summary
Microsoft has launched a new class of artificial intelligence (AI) models able to process text, images and speech simultaneously, requiring less computing power than existing systems.
Called Phi-4 models, they are designed to run on standard hardware or at the “edge”, on devices directly rather than in cloud data centres, and can be used to develop innovative, context-aware applications, says Weizhu Chen, vice president of generative AI at Microsoft.
Designed to be efficient, the models have a novel “mixture of LoRAs” technique, which enables them to handle various input types without sacrificing accuracy.
They can be used to enhance efficiency and accuracy, according to Steve Frederickson, head of product at AI answer engine Capacity, which has already leveraged the Phi family to enhance its platform’s efficiency and accuracy, achieving a 4.2x cost saving compared to competing workflows.