Llama is a family of language models whose weights are published, so they can be run on your own hardware instead of only through an API. It solves the problem that most capable models are reachable only as a hosted service, which rules them out when data cannot leave your network or when per-token cost dominates. It exists because releasing weights turned out to build an ecosystem of fine-tunes and tooling around one architecture, and it is the model most local inference tools target first.
Phi is a family of small language models trained on a deliberately narrow, high-quality corpus of textbooks and exercises rather than on a web-scale crawl. It solves the assumption that capability requires scale, showing that data selection can substitute for a large part of it. It exists because it reframes the question from how much data to what kind of data, and the answer is useful when the budget is a single GPU.
