New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Update transpose doc #6072
base: main
Are you sure you want to change the base?
Update transpose doc #6072
Conversation
Signed-off-by: Ganesan Ramalingam <grama@microsoft.com>
Would it be a good idea to address #4611 in the same PR? |
Maybe. But at least some of those questions require some investigation ... it is easy enough to say 0D and 1D inputs should be supported (the op is well-defined for these cases). But that doesn't mean that a specific implementation will support them ... only if we test it out, will we realize if some assumptions are being made. |
Signed-off-by: Ganesan Ramalingam <grama@microsoft.com>
Added point about perm being optional. |
Codecov ReportAll modified and coverable lines are covered by tests ✅
Additional details and impacted files@@ Coverage Diff @@
## main #6072 +/- ##
==========================================
+ Coverage 56.97% 56.99% +0.01%
==========================================
Files 506 506
Lines 30483 30489 +6
Branches 4593 4593
==========================================
+ Hits 17369 17378 +9
+ Misses 12285 12283 -2
+ Partials 829 828 -1 ☔ View full report in Codecov by Sentry. |
As for the question about 0D input tensors (scalars): it looks like both ONNX shape inference as well as its reference implementation need to be extended/fixed to handle this case. |
Description
Clarify the permutation specification of the Transpose op.
Motivation and Context
It is not clear from the spec whether the i-th input axis corresponds to the perm[i]-th output axis or whether the i-th output axis corresponds to perm[i]-th input axis. Make this clear.