Skip to content

Generate shape-independent graph if any input dimension < 2 #6581

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.

Sign up for GitHub

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

Merged
wschin merged 3 commits into from
Feb 10, 2021
Merged

Generate shape-independent graph if any input dimension < 2 #6581

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
4d8707d
Throw for non-supported case
wschin Feb 5, 2021
6c3a538
Not to go to shape-dependent branch when seeing unsupported shapes
wschin Feb 5, 2021
02350e0
Merge remote-tracking branch 'public/master' into wechi/throw-matmul-…
wschin Feb 10, 2021
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
3 changes: 2 additions & 1 deletion orttraining/orttraining/core/graph/gradient_builder.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -194,7 +194,8 @@ IMPLEMENT_GRADIENT_BUILDER(GetMatMulGradient) {
}
};

if (A_has_shape && B_has_shape && Y_has_shape) {
if (A_has_shape && B_has_shape && Y_has_shape &&
A_shape.size() >= 2 && B_shape.size() >= 2) {
Copy link
Contributor

@SherlockNoMad SherlockNoMad Feb 5, 2021

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Matmul schema only allow inputs with dim>=2.
I thought this is an implicit prerequisite? Are you seeing inputs with less then 2 dim?

Copy link
Contributor

@SherlockNoMad SherlockNoMad Feb 5, 2021 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I saw the discussion on team's channel, and understand now this is caused by a torch bug/"feature".
This fix is harmless.

Copy link
Member

@yuslepukhin yuslepukhin Feb 8, 2021

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Out MatMul allows vectors.

In reply to: 571201490 [](ancestors = 571201490)

std::vector<AttributeProto> shared_attributes;
shared_attributes.push_back(MakeAttribute("beta", float(0)));
AttributeProto transpose_first_input = MakeAttribute("transA", int64_t(1));
Expand Down