[SYSTEMDS-3303] : NN Builtin: Attention Layer #1625
Closed
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
…nd try the attention layer.
modified attention_layer toy example
…ction documentation regarding inputs and outputs for forward and backward, prototype of backward pass
…tion layer, fixed backward pass for attention layer
Baunsgaard
reviewed
Jun 6, 2022
Contributor
There was a problem hiding this comment.
Overall i like this PR,
There are some things that needs to be addressed:
- For performance (removing for loop)
- For consistency add the arguments for the attention matrix and gradients for forward and backwards call to make the functions like the other operations.
- The example does not verify if the attention behaves correctly in case of different dimensions for query and value, and it would be nice if we could have this.
- I do not understand the test, and how that verifies if the method works but i think some comments in the tests would help solve that.
- Remove the data from the PR, and add a download script and change .gitignore to ignore downloaded file.
best regards
Sebastian
| } | ||
| } | ||
|
|
||
| attention = function() { |
Contributor
There was a problem hiding this comment.
I do not understand what this method is testing, maybe a comment would help.
Contributor
Author
There was a problem hiding this comment.
This test is to verify the gradient of the backward pass numerically.
The comments are just copied and modified from the other test cases in this file.
Contributor
There was a problem hiding this comment.
The intention of me commenting on it was, that you add a comment in code. When you add it here only i get the help from you.
Contributor
Author
|
@Baunsgaard Thanks for the feedback. We'll address the feedback shortly. |
…the AttentionExample.dml file.
Contributor
Author
|
@Baunsgaard i think we addressed all issues, with the exception of the for loop in the layer. |
A-Postl
changed the title
A-Postl
changed the title
…ttentionExample.sh script can be found.
Baunsgaard
reviewed
Jun 23, 2022
Contributor
Baunsgaard
left a comment
Baunsgaard
left a comment
There was a problem hiding this comment.
Thanks for addressing my comments,
I will take it from here.
Baunsgaard
pushed a commit
to Baunsgaard/systemds
that referenced
this pull request
Aug 17, 2022
This commit adds a new Neural network builtin layer for attention. AMLS project SS2022 Co-authored-by: Anton Postl <anton.postl@student.tugraz.at> Co-authored-by: Stefan Schörkmeier <s.schoerkmeier@student.tugraz.at> Closes apache#1625
Baunsgaard
pushed a commit
to Baunsgaard/systemds
that referenced
this pull request
Aug 17, 2022
This commit adds a new neural network builtin layer for attention. AMLS project SS2022 Closes apache#1625 Co-authored-by: Anton Postl <anton.postl@student.tugraz.at> Co-authored-by: Stefan Schörkmeier <s.schoerkmeier@student.tugraz.at>
Contributor
|
Closing for merging. |
fathollahzadeh
pushed a commit
to fathollahzadeh/systemds
that referenced
this pull request
Dec 7, 2022
This commit adds a new neural network builtin layer for attention. AMLS project SS2022 Closes apache#1625 Closes apache#1679 Co-authored-by: Anton Postl <anton.postl@student.tugraz.at> Co-authored-by: Stefan Schörkmeier <s.schoerkmeier@student.tugraz.at>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
5 participants
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.
This PR implements an attention layer using scaled dot-product as the alignment score function,
as well as an example showing the usage of the attention layer (as self-attention) in combination with an LSTM layer on
a text classification problem on real data.