Linearizing Attention | Towards Data Science

Breaking the Quadratic Barrier: Modern Alternatives to Softmax Attention