Fetching the paper…

Implicit Regularization of Gradient Flow on One-Layer Softmax Attention · Around