Remove detaches to enable gradient-based force EquiformerV2 models. #779
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
Codecov ReportAll modified and coverable lines are covered by tests ✅
|
mshuaibii
reviewed
Jul 22, 2024
lbluque
previously approved these changes
Jul 22, 2024
There was a problem hiding this comment.
lgtm.
Agreed with @mshuaibii that making sure direct force models are consistent. In addition it might be worth to write a simple unit-test to make sure the model is indeed properly differentiable.
Comment on lines
+57
to
+65
| output = torch.zeros_like(edge_rot_mat) | ||
| mask = (yprod > -0.9999) & (yprod < 0.9999) | ||
| output[mask] = edge_rot_mat[mask] | ||
| output[~mask, 0, :] = edge_rot_mat[~mask, 0, :] | ||
| output[~mask, 2, :] = edge_rot_mat[~mask, 2, :] | ||
| output[yprod > 0.9999, 1, :] = edge_rot_mat.new_tensor([[0., 1., 0.]]) | ||
| output[yprod < -0.9999, 1, :] = edge_rot_mat.new_tensor([[0., -1., 0.]]) | ||
|
|
||
| return output |
There was a problem hiding this comment.
nit, can this section be replaced with?
edge_rot_mat[yprod > 0.9999, 1, :] = edge_rot_mat.new_tensor([[0., 1., 0.]])
edge_rot_mat[yprod < -0.9999, 1, :] = edge_rot_mat.new_tensor([[0., -1., 0.]])
return edge_rot_mat
|
This PR has been marked as stale because it has been open for 30 days with no activity. |
|
Egrad models will be in dev-mode for now. |
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Outdated suggestions cannot be applied.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.
Remove the detaches in obtaining rotation matrices and wignerD matrices for correct gradients of model outputs with regard to atomic positions.
Naively remove the detaches result in instability due to gradient explosion for y-axis rotation of ~0 or ~180 degree.
In this fix, we instead do not rotate / rotate 180 degree for edges that are already aligned with the y-axis. This fix enables correct gradients of model outputs with regard to the input atom positions and thus enable training energy-conserving gradient-based force EquiformerV2 models.