Fetching the paper…

Multi-Layer Transformers Gradient Can be Approximated in Almost Linear Time · Around