Skip to main content

Phi-2: When Less Is More Powerful

·153 words·1 min

Mistral-7B is cool but you know what’s cooler? A more powerful model in just 1/3rd of the size!

Welcome to Phi-2.

This is something our team at Microsoft Research had been tirelessly working on and now we have more numbers comparing with Llama-7B, 13B, 70B and Gemini Nano. 👇

Phi-2 is best in the class model beating models 3X of it’s size and in many cases even much much larger. Phi-2 especially shines in Math and Coding:

https://pbs.twimg.com/media/GBODsdIa8AA7Zox.png

Want to give Phi-2 a try? Here is how:

  1. Go to http://portal.azure.com, create an account if you don’t have one

  2. Go here: http://aka.ms/phi2az

Our Phi-2 efforts proves that we had been wasting enormous amount of compute on rather ineffective training data. Throwing a kitchen sink at model has a big price tag and lower quality.

More examples and discussions: https://x.com/SebastienBubeck/status/1734612826176794958

Blog Post: https://microsoft.com/en-us/research/blog/phi-2-the-surprising-power-of-small-language-models/

By popular demand, Phi-2 is now directly available on Hugging Face!!

https://x.com/SebastienBubeck/status/1735050282210615431

Discussion