Demo: Phi llamafile - Llama Tutorial

- [Presenter] Phi might be one of the most canonical examples of a small language model that has been created recently. And we can take a look here that it is built by Microsoft Research, it's got 2.7 billion parameters and it has some good characteristics here. And one of the things that we can take a look at here is that it is using this textbook quality data. So that's really one of the things that's different than these really large models is that there is a focus on extremely high quality data. So this could also show that maybe in the future, this is the direction AI goes, is they partner with content creators and they specialize in training models on extremely high data. This could really change things up quite a bit, because one of the things we can do if we look into the benchmarking here is that we can see that even though there are a much smaller amount of parameters, 2.7 billion parameters here, that it is able to beat other large language models on things like reasoning…