Closed inexorabletash closed 4 months ago
It seemed like this was the only actionable work for #590
I was pondering adding informative notes to the algorithms e.g. explaining when/where bidirectional/unidirectional broadcasting should be used, but inspiration didn't strike. Suggestions welcome!
@a-sully and @fdwr can you take a look?
... and if we want more text about broadcasting, I sketched something out in https://github.com/webmachinelearning/webnn/issues/378#issuecomment-2253655000 - I could pull that into this PR if desired.
LGTM 👍 I especially like s/broadasting the shapes/broadcasting
Just say "broadcast" not "broadcasted"; Merriam-Webster says the former is preferable.
Make prelu() bidirectionally broadcast instead of unidirectionally broadcast, as it's not an obvious unidirectional case.
Make preamble more consistent across operations in calling out broadcasting and stating the requirements for input tensors.
Shorten algorithm link, e.g. from "...broadasting the shapes A's shape and B's shape" to just "...broadcasting A's shape and B's shape"
Resolves #590