Bug fix: fix setting different learning rates between backbone and main model in ACT policy #280
+32
−2
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
alexander-soare
approved these changes
Jun 18, 2024
There was a problem hiding this comment.
Thanks for this @thomwolf and for adding the test to be sure. I'm wondering if it also makes sense to put an assert in train.py (assert that the param groups have non-zero length, and maybe even that their lengths sum up to the total for the model). I most certainly will be making a habit of this from now on as it is such a sneaky and silent bug.
|
Thanks. I think it's probably quite fine now since we only have the backbone in this script (maybe to be added in other scripts indeed) |
|
Feel free to merge whenever is convenient for you |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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Outdated suggestions cannot be applied.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.
What this does
This pull request fix a bug where we were not identifying the backbone in the filtering of the parameters when specifying the learning rate for the backbone and for the main model.
As a result we were always assigning the main model LR for the back bone instead of the backbone specific LR,
lr_backbone.How it was tested
./tests/test_policies.py::test_act_backbone_lr.How to checkout & try? (for the reviewer)
It can be tested with
pytest ./tests/test_policies.py::test_act_backbone_lrThis change is