Skip to content
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

LSTM/GRU/RNN Sequences : support for seq_lengths input #2788

Merged

Conversation

itikhono
Copy link
Contributor

@itikhono itikhono commented Oct 23, 2020

  • Sequences to TI transformations
  • Adjust reference implementations of GRU/LSTM/RNN Seqs
  • fix ngraph python API
  • IR Reader unit tests
  • enable ONNX importer unit tests
  • fix TensorIterator ref impl and tests

@itikhono itikhono requested a review from iefode November 12, 2020 09:22
@itikhono itikhono marked this pull request as ready for review November 12, 2020 10:15
@itikhono itikhono requested review from a team November 12, 2020 10:15
@itikhono itikhono requested review from a team as code owners November 12, 2020 10:15
@itikhono itikhono requested review from a team and AlexPeskov November 12, 2020 10:15
@itikhono itikhono added this to the 2021.2 milestone Nov 12, 2020
@itikhono itikhono added category: Core OpenVINO Core (aka ngraph) category: IE Tests OpenVINO Test: plugins and common labels Nov 12, 2020
@itikhono itikhono requested a review from ilyachur November 13, 2020 05:58
body_params.push_back(aggregated_Y_h_body_param);

// Create mask node deciding whether or not to mask batch data.
ngraph::Output<ngraph::Node> batch_seq_length = ngraph::builder::opset1::legacy_broadcast_for_binary_operation(
Copy link
Contributor

Choose a reason for hiding this comment

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

If you create opset5 operations during transformation then you can't create opset1.

Copy link
Contributor Author

@itikhono itikhono Nov 14, 2020

Choose a reason for hiding this comment

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

@GlebKazantaev opset5 ops are aliases to opset1, for all ops in this legacy_broadcast_for_binary_operation function.
I don't want to duplicate the code of this function and don't want to upgrade the function to opset5 due to there are other places of use of this function.

This is a general problem in our code: we cannot create helper functions with a hard-coded version of the opset. It is a more complicated issue than using this function. I will open a question in our morning meeting.

Copy link
Contributor Author

Choose a reason for hiding this comment

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

And we also allow MO to generate ops from different opsets

@itikhono itikhono force-pushed the itikhono/sequences_to_tensor_iterator branch from 680e2cc to 5014cf2 Compare November 14, 2020 09:11
@ilyachur
Copy link
Contributor

@itikhono please resolve conflicts.

@ilyachur ilyachur merged commit b45e1a2 into openvinotoolkit:master Nov 17, 2020
mryzhov pushed a commit to mryzhov/openvino that referenced this pull request Nov 17, 2020
…it#2788)

* sequences to ti transformations, support for seq_lengths input, update reference implemetations, add new tests

* fix python api, update sequences to ti transformation

* fix sequences to ti transformation

* Update sequences to TI transformation: fix reverse sequence support

* update single layer tests, fix TI reference impl, fix Sequences to TI transformations

* ngraph code style

* fix build

* fix ngraph python api

* resolver review comments, refactoring

* Resolve review remarks

* delete xfail
@openvinotoolkit openvinotoolkit deleted a comment from itikhono Nov 17, 2020
mryzhov pushed a commit to mryzhov/openvino that referenced this pull request Dec 16, 2020
…it#2788)

* sequences to ti transformations, support for seq_lengths input, update reference implemetations, add new tests

* fix python api, update sequences to ti transformation

* fix sequences to ti transformation

* Update sequences to TI transformation: fix reverse sequence support

* update single layer tests, fix TI reference impl, fix Sequences to TI transformations

* ngraph code style

* fix build

* fix ngraph python api

* resolver review comments, refactoring

* Resolve review remarks

* delete xfail
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
category: Core OpenVINO Core (aka ngraph) category: IE Tests OpenVINO Test: plugins and common
Projects
None yet
Development

Successfully merging this pull request may close these issues.

6 participants