Microsoft announces new artificial intelligence Phi-2, which promises to be more efficient than Llama-2

Microsoft has released Phi-2, its new small language model (SLM). Artificial intelligence (AI) is an evolution of Phi-1 and, according to the company, Can outperform teams 25 times its size in performance tests.

Phi-2 is described by Microsoft as the “ideal playground” for researchers. It can help develop a variety of tasks, all within Azure AI Studio.

Today we’re sharing our teams’ latest contributions, Phi-2 and Promptbase.

Phi-2 outperforms other existing small language models but is small enough to run on a laptop or mobile device. https://t.co/wLhUeRsByL

—Microsoft Research (@MSFTResearch) December 12, 2023

Unlike traditional language models called LLM, Phi-2 has fewer parameters. This provides “lighter” but still more efficient data processing than Meta’s Mistral and Llama-2 in benchmark tests.

According to Microsoft, Phi-2 is a predictive word association model for content creation like ChatGPT. It was trained for 14 days on 1.4 trillion tokens obtained from synthetic data taken from the internet.

Phi-2 can even outperform Gemini, Google’s most advanced language model. Microsoft points out that in some scenarios, the project itself, as well as Google’s artificial intelligence, can solve problems.

Because it is a small language model and calculates fewer variables to produce results, Phi-2 requires less computing power to run. This makes the platform cheaper to use, which is necessary for simpler applications.

Phi-2 is for research purposes only

However, Phi-2 is not available for commercial or public use. The tool is intended for exclusive use by researchers, so you won’t see it running in Microsoft applications or services (at least not yet).