Shulou(Shulou.com)11/24 Report--

CTOnews.com, November 17, Microsoft Ignite 2023 conference has begun, Microsoft released its smallest language model Phi-2, a total of 2.7 billion parameters, compared with the previous version has been significantly improved.

CTOnews.com Note: Microsoft released Phi-1 in June this year, with only 1.3 billion parameters, suitable for QA Q & A, chat format and code scenarios. The model is trained entirely based on high-quality data and performs 10 times better than similar models in benchmark tests.

Microsoft released an updated Phi-1.5 version in September, also with 1.3 billion parameters, which can write poems, e-mails and stories, as well as summary texts. In the benchmarking of common sense, language understanding, and reasoning, the model can keep up with models with up to 10 billion parameters in some areas.

Microsoft's updated Phi-2 now has 2.7 billion parameters, which is twice the size of the previous version, but it is still much smaller than other mainstream language models.

Microsoft says Phi-2 has shown significant improvements in logical reasoning and security. With the right fine tuning and customization, small language models are powerful tools for cloud and edge applications.

Sebastien Bubeck, head of the Microsoft Research machine learning foundation team, posted a screenshot of the "MT-Bench" benchmark on Twitter, testing the actual functionality of large and small language models with powerful language models such as GPT-4.

According to the results, Phi-2 is better than Meta's Llama-2-7B model in some aspects. A chat version of Phi-2 is also in the pipeline and may address some of the existing pain points of the model in these areas.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.