Microsoft Research reposted this
🚀 Phi-4 is here! A small language model that performs as well as (and often better than) large models on certain types of complex reasoning tasks such as math. Useful for us in @MSFTResearch, and available now for all researcher on the Azure AI Foundry! https://aka.ms/phi4blog
Will there also be any multi-modal variants of phi-4? 🙂
Very nice! Hopefully it can be used by the broader community soon. Come share what you build and learn with us in the AI Agents group on linkedin: https://www.linkedin.com/groups/6672014
(a) It's a shame the license is a limited research one (b) It's fascinating that phi-4, a 14B parameter model, actually outperforms GPT-4 (its teacher) on high-level STEM tasks like graduate physics and math competition problems. Seems thanks to the specialized synthetic data and training methods mentioned in the paper.
Just yesterday I asked Copilot and I said wow these characteristics would be very good, I thought I was hallucinating, now I see that I am not. Microsoft must set a firm strategy and of course in parallel to that of openai with Phi
Phi-4 proves that bigger isn’t always better, this 14B parameter small language model is rewriting the rules of complex reasoning, especially in math. Excited to see how it accelerates innovation on Azure AI Foundry and beyond ..
Finally a model dedicated to arithmetic
Great news can't wait to test ist on my local machine
Microsoft’s focus on smaller, more efficient models could have significant implications for businesses, particularly for smaller enterprises that may not have the resources to support large-scale AI systems. The release of Phi-4 is also being carefully monitored for safety and ethical considerations, with features designed to prevent misuse.
Love the work your team is doing.
Chief Architect genAI, AI, HC & LS @ Progress | M.Eng.
1wWow totally punching above its weight.