TensorTonic
Problems

Implement Scaled Dot-Product Attention

Implement batched scaled dot-product attention for self- and cross-attention with optional masks and stable softmax.