makemore

Important references

• PyTorch broadcasting rules — read this carefully before doing tensor arithmetic across dimensions of different shapes (e.g. row/column normalization). Getting keepdim wrong here causes silent, hard-to-spot bugs.