Microsoft launches Phi-2, a small language AI model that outperforms Llama 2 and Mistral 7B

Are You ready has bring more awareness has your brand? Consider become A sponsor For THE AI Impact Tour. Learn more about THE opportunities here.

THE fast pace of generative AI news And advertisement is not it to slow down down, even as We reach THE final stretches of 2023 And THE traditional winter vacation calm period.

Just take A look has Microsoft Research, THE blue sky division of THE software giant, which Today announcement THE release of It is Phi-2 little language model (LML), A text to text AI program that East "little enough has run on A laptop Or mobile device," according to has A job on X.

HAS THE even time, Phi-2 with It is 2.7 billion settings (Connections between artificial neurons) is bragging performance that East comparable has other, a lot bigger models including The meta Lama 2-7B with It is 7 billion settings And even Mistral-7B, another 7 billion setting model.

Chart comparing Microsoft Research Phi-2 model has other leading open source And farm source models. Credit: Microsoft Search

Microsoft researchers Also note In their Blog job on THE Phi-2 release that he surpasses that of Google brand new Gemini Nano 2 model despite he having half A billion more settings, And book less "toxicity" And bias In It is answers that Lama 2.

V.B. Event

THE AI Impact Visit

Connect with THE business AI community has VentureBeat AI Impact Tour future has A city close you!

Learn More

Microsoft Also I could not resist socket A little dig has Google...