1 post on self-attention, newest first.
How self-attention works in Transformers: queries, keys and values, softmax weights, a worked pronoun example, multi-head attention and the O(n²) cost.