BooksLecture notes
Efficient Transformer Inference: Lecture notes
Lecture notes from a graduate course on making transformer inference fast and small: attention variants, caching, speculative decoding, quantisation, batching and scheduling. Includes problem sets and a reading list.