Skip to content

Conversation

@RandySheriffH
Copy link
Collaborator

@RandySheriffH RandySheriffH commented May 7, 2020

A tentative fix with trivial change to match LSTM for version between 1.15.0 and 2.x.
Addressing issue #861

@RandySheriffH RandySheriffH requested review from guschmue and jignparm May 7, 2020 20:14
@RandySheriffH RandySheriffH changed the title Fix LSTM pattern matching for 1.15.0 <= tf < 2.x Fix LSTM pattern matching for version between 1.15.0 and 2.x. May 7, 2020
]),
]),
])

Copy link
Contributor

Choose a reason for hiding this comment

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

Any UT that can be re/activated to check if this new pattern is working as expected? -- or are the current tests already sufficient?

Copy link
Collaborator Author

@RandySheriffH RandySheriffH May 7, 2020

Choose a reason for hiding this comment

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

Current cases in test_lstm.py already covering it.
I have enabled check_lstm_count(...) in common.py to capture issues.

]),
])

xc_pattern_optimized = \
Copy link
Contributor

Choose a reason for hiding this comment

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

why is the name _optimized ?

Copy link
Collaborator Author

Choose a reason for hiding this comment

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

@guschmue :
My understanding is for tf >= 1.15.0 when bias is all zero, "BiasAdd" will be optimized to "Identity" by tf optimizer, hence the name.

@RandySheriffH RandySheriffH merged commit 9ec80aa into onnx:master May 8, 2020
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants