Be able to work out, from a model's size and a machine's published numbers, how fast it can answer, where the time is actually going, and which of the available settings would change the answer
What It Costs to Run a Model

Running a trained model is an arithmetic problem with a small number of terms: how many bits each number takes, how many bytes must be moved for every word produced, how much the kept conversation adds, and how many requests are being served at once.
8 lessons, written and corrected before you arrived. Reading them here needs no account. The first reads the whole way through; the others open and then stop, because a page nobody owns cannot tell who is reading it. Starting the course gives you your own copy, where every idea has problems standing under it and you can ask about any sentence.