Clip grad norm support for dtensors - #4219
Merged
Merged
Conversation
|
The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update. |
SunMarc
reviewed
Sep 8, 2026
Member
There was a problem hiding this comment.
Thanks ! This was kind of tackled in transformers trainer but happy to have it here also ! Can you just check the impl differences and maybe align if necessary ? huggingface/transformers#48208
Comment on lines
+3022
to
+3026
| if not any(isinstance(p.grad, DTensor) for p in parameters if p.grad is not None): | ||
| return torch.nn.utils.clip_grad_norm_(parameters, max_norm, norm_type=norm_type) | ||
|
|
||
| dtensor_params = [p for p in parameters if p.grad is not None and isinstance(p.grad, DTensor)] | ||
| plain_params = [p for p in parameters if p.grad is not None and not isinstance(p.grad, DTensor)] |
Member
There was a problem hiding this comment.
maybe create the dtensor_params once and check for it ?
Comment on lines
+3014
to
+3031
| parameters = list(parameters) | ||
|
|
||
| is_dtensor_available = torch.distributed.is_available() and is_torch_version(">=", DTENSOR_PYTORCH_VERSION) | ||
| if not is_dtensor_available: | ||
| return torch.nn.utils.clip_grad_norm_(parameters, max_norm, norm_type=norm_type) | ||
|
|
||
| from torch.distributed.tensor import DTensor | ||
|
|
||
| if not any(isinstance(p.grad, DTensor) for p in parameters if p.grad is not None): | ||
| return torch.nn.utils.clip_grad_norm_(parameters, max_norm, norm_type=norm_type) | ||
|
|
||
| dtensor_params = [p for p in parameters if p.grad is not None and isinstance(p.grad, DTensor)] | ||
| plain_params = [p for p in parameters if p.grad is not None and not isinstance(p.grad, DTensor)] | ||
| group_norms = [ | ||
| torch.nn.utils.get_total_norm([p.grad for p in group], norm_type) | ||
| for group in (dtensor_params, plain_params) | ||
| if group | ||
| ] |
Member
There was a problem hiding this comment.
this is a specific case when we have mixed dtensors as we have here no ? huggingface/transformers#48208. maybe have a check for mixed grad + a seperate function to calculate the correct grad. otherwise return torch.nn.utils.clip_grads_with_norm_
SunMarc
approved these changes
Sep 8, 2026
Comment on lines
+2971
to
+2972
| if len(mesh_groups) + bool(plain_params) <= 1: | ||
| return torch.nn.utils.clip_grad_norm_(parameters, max_norm, norm_type=norm_type) |
Co-authored-by: Marc Sun <57196510+SunMarc@users.noreply.github.com>
SunMarc
reviewed
Sep 8, 2026
SunMarc
reviewed
Sep 8, 2026
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 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.
What does this PR do?
Seperate the plain tensors from the dtensors, otherwise the foreach op in
torch.nn.utils.clip_grad_norm_fails.