Phi-4: The Most Powerful smol Model!
Are you ready for an early Christmas present from our team at Microsoft Research?
Introducing the most powerful smol model ever built in the world!
Welcome to Phi-4! 👇
Personally, since I started working on Phi project at its inception, this is my most favorite model that we have ever shipped.
Remember prompts that many frontier models including o1-preview struggled with? Phi-4 gave correct answer super fast!
https://x.com/sytelus/status/1853188232219128220
Phi-4 achieves this by pushing the art of synthetic data even further to induce reasoning abilities along with new advancements in post-training.
If you have been using Llama 3.x etc for reasoning tasks, you owe it to yourself to try out Phi-4!
This is the work of super hard working team including @EldanRonen @BehlHarkirat @mojan_jp @marah @marah_i_abdin @weishliu @rosaguga @OlliSaarikivi @suriyagnskr @SebastienBubeck and many others.
Of course, none of this would have been possible without support from our amazing…
Paper: https://arxiv.org/abs/2412.08905
Model is available now to try in Azure: https://ai.azure.com/explore/models/Phi-4/version/1/registry/azureml
What’s better than tokens? Tokens with logits!!
We hope you will have as much fun playing and creating with this tiny beast of a model as we had building it.
Happy holiday chilln’ for ya all🎄🎁.
Oh… and one more thing! if you love to push model capabilities and want to work with us, check out our summer internship position. I am at #NeurIPS and let me know if you had like to chat more about this position.
https://x.com/sytelus/status/1866639121097298372
Tagging more people! @NguynTu24128917 @dingli_yu @xinw_ai @yw_yuewu @YiZhangZZZ @AdilSlm @_cyrilzhang
Also, check out our blog post at https://aka.ms/phi4blog for some great examples of reasoning tasks where Phi-4 excels!
Way better summary than I can do by one and only @rohanpaul_ai. https://x.com/rohanpaul_ai/status/1867426966305222929