Do not aggregate the losses since last log step #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
facebook-github-bot
added
the
CLA Signed
carmocca
commented
Jan 7, 2025
Comment on lines
+340
to
+341
| utils.dist_mean(loss, world_mesh["dp_cp"]), | ||
| utils.dist_max(loss, world_mesh["dp_cp"]), |
There was a problem hiding this comment.
These use functional collectives under the hood so there shouldn't be issues with passing in the same tensor reference
| @@ -34,12 +34,12 @@ def get_device_info(): | |||
| device_type, device_module = get_device_info() | |||
|
|
|||
|
|
|||
| def dist_max(x: Union[int, float], mesh: DeviceMesh) -> float: | |||
| def dist_max(x: Union[int, float, torch.Tensor], mesh: DeviceMesh) -> float: | |||
| tensor = torch.tensor(x).to(device_type) | |||
There was a problem hiding this comment.
AFAIK these will result in no-ops if a tensor of with the same device type is passed.
tianyu-l
reviewed
Jan 7, 2025
Comment on lines
+340
to
+341
| utils.dist_mean(loss, world_mesh["dp_cp"]), | ||
| utils.dist_max(loss, world_mesh["dp_cp"]), |
There was a problem hiding this comment.
It seems with Tensor Parallel, the loss is a DTensor, which doesn't support functional collectives. Also we should not require gradients on this all-reduce.
Maybe it's still fine to do .item() outside as before? or use detach and full_tensor() ?
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Outdated suggestions cannot be applied.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.
Fixes #763