I really appreciate this type of articles. I feel like a lot of knowledge in LLM training and inference is locked inside the heads of practitioners. Similar to compiler engineers before.
To work in LLM training/inference you’re expected to know this stuff but to know this stuff you need to be working in the space.