Back to blog
Notes
A shape mismatch is usually a missing mental model, not just a typo.
Good ML notes explain where the tensor dimensions go.
If the abstraction hides the performance cost, open the abstraction.