As I said, the attention step is O(n^2) per token sequentially and O(n^2) for the entire sequence when calculating the entire sequence in parallel, where n is t

As I said, the attention step is O(n^2) per token sequentially and O(n^2) for the entire sequence when calculating the entire sequence in parallel, where n is the length of the sequence.