Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
What does this PR do?
This PR refactors ProphetNet similar to Bart in that it moves the time dimension to be always at the 2nd place and the batch dimensions always in the first place. Also, the cache is refactored to consists of tuples instead of a dict.
The model is thereby very much aligned with Bart (I cannot really add any " # Copied from" statements though because the weight names are different).
The PR is in spirit very similar to #8900.
I've verified that all slow tests pass. In the next step, I want to make a short notebook, verifying that ProphetNet can be trained since there have been some issues on training: #9804
Benchmarking
The PR doesn't change compute or memory complexity:
On this PR:
on master: