This repository has been archived by the owner on Nov 3, 2023. It is now read-only.
Extend BERT-based classification with customized layers #4553
Conversation
This file contains 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
klshuster
reviewed
May 18, 2022
| @@ -90,20 +99,6 @@ def add_cmdline_args( | |||
| """ | |||
| super().add_cmdline_args(parser, partial_opt=partial_opt) | |||
| parser = parser.add_argument_group("BERT Classifier Arguments") | |||
| parser.add_argument( | |||
There was a problem hiding this comment.
was this option just never used?
There was a problem hiding this comment.
Right. I actually also found it in BertWrapper's add_common_args function, also never used, so I'll remove it from there
|
|
||
| if ind < len(dimensions): | ||
| raise Exception( | ||
| "Output layer's dimension does not match number of classes. Found {dimensions[ind][1]}, expected {output_dimension}" |
There was a problem hiding this comment.
nit: think you're missing f"" string here
| "Output layer's dimension does not match number of classes. Found {dimensions[ind][1]}, expected {output_dimension}" | ||
| ) | ||
| raise Exception( | ||
| "Output layer's dimension does not match number of classes. Found {prev_dimension}, expected {output_dimension}" |
parlai/agents/bert_ranker/helpers.py
Outdated
| aggregation="first", | ||
| bert_model: BertModel, | ||
| output_dim: int = -1, | ||
| classifier_layer: torch.nn.Module = None, |
There was a problem hiding this comment.
nit: could you please move this arg to be the last one? so that prior calls to this __init__ don't fail
There was a problem hiding this comment.
fixed issues and added tests
klshuster
approved these changes
May 19, 2022
|
I have changed the torch version in CircleCI config to make it work since CR was approved, so re-requesting approval for this change |
kushalarora
pushed a commit
that referenced
this pull request
Jun 15, 2022
* Extend BERT-based classification with customized layers * fix bugs and add tests * increase lr to improve training stability * upgrading torch version * adjusting loss value
Sign up for free
to subscribe to this conversation on GitHub.
Already have an account?
Sign in.
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.
Patch description
Added functionality to specify custom decoder layers for BERT-based classification. Code is a modification of existing in external ParlAI functions.
Testing steps
Other information