New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
[Gradient Compression] Add an index field to GradBucket for PowerSGD #48757
Conversation
Add an index field to GradBucekt, so error_dict is keyed by this index instead of the hashcode of input tensor. Howevever, sometimes the buckets can be rebuilt in the forward pass. In this case, the shape of the bucket with the same index will not be consistent with the one in the previous iteration, and hence the error tensor will be re--initialized as a zero tensor of the new shape. Original PR issue: Investigate Applying PowerSGD to Communication Hook for Gradient Compression #47202 Differential Revision: [D25288496](https://our.internmc.facebook.com/intern/diff/D25288496/) [ghstack-poisoned]
Add an index field to GradBucekt, so error_dict is keyed by this index instead of the hashcode of input tensor. Howevever, sometimes the buckets can be rebuilt in the forward pass. In this case, the shape of the bucket with the same index will not be consistent with the one in the previous iteration, and hence the error tensor will be re--initialized as a zero tensor of the new shape. Original PR issue: Investigate Applying PowerSGD to Communication Hook for Gradient Compression #47202 Differential Revision: [D25288496](https://our.internmc.facebook.com/intern/diff/D25288496/) ghstack-source-id: 117719173 Pull Request resolved: #48757
💊 CI failures summary and remediationsAs of commit c05592d (more details on the Dr. CI page): ✅ None of the CI failures appear to be your fault 💚
🚧 1 ongoing upstream failure:These were probably caused by upstream breakages that are not fixed yet:
This comment was automatically generated by Dr. CI (expand for details).Follow this link to opt-out of these comments for your Pull Requests.Please report bugs/suggestions on the GitHub issue tracker or post in the (internal) Dr. CI Users group. This comment has been revised 26 times. |
…r PowerSGD" Add an index field to GradBucekt, so error_dict is keyed by this index instead of the hashcode of input tensor. Howevever, sometimes the buckets can be rebuilt in the forward pass. In this case, the shape of the bucket with the same index will not be consistent with the one in the previous iteration, and hence the error tensor will be re--initialized as a zero tensor of the new shape. Original PR issue: Investigate Applying PowerSGD to Communication Hook for Gradient Compression #47202 Differential Revision: [D25288496](https://our.internmc.facebook.com/intern/diff/D25288496/) [ghstack-poisoned]
Pull Request resolved: #48757 Add an index field to GradBucekt, so error_dict is keyed by this index instead of the hashcode of input tensor. Howevever, sometimes the buckets can be rebuilt in the forward pass. In this case, the shape of the bucket with the same index will not be consistent with the one in the previous iteration, and hence the error tensor will be re--initialized as a zero tensor of the new shape. Original PR issue: Investigate Applying PowerSGD to Communication Hook for Gradient Compression #47202 ghstack-source-id: 117748471 Differential Revision: [D25288496](https://our.internmc.facebook.com/intern/diff/D25288496/)
…r PowerSGD" Add an index field to GradBucekt, so error_dict is keyed by this index instead of the hashcode of input tensor. Howevever, sometimes the buckets can be rebuilt in the forward pass. In this case, the shape of the bucket with the same index will not be consistent with the one in the previous iteration, and hence the error tensor will be re--initialized as a zero tensor of the new shape. Original PR issue: Investigate Applying PowerSGD to Communication Hook for Gradient Compression #47202 Differential Revision: [D25288496](https://our.internmc.facebook.com/intern/diff/D25288496/) [ghstack-poisoned]
Pull Request resolved: #48757 Add an index field to GradBucekt, so error_dict is keyed by this index instead of the hashcode of input tensor. The replacement will be done in a separate diff, as the definition of this new method somehow couldn't be recognized in the OSS version. Original PR issue: Investigate Applying PowerSGD to Communication Hook for Gradient Compression #47202 ghstack-source-id: 117911015 Differential Revision: [D25288496](https://our.internmc.facebook.com/intern/diff/D25288496/)
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
LGTM, please ensure CI is green
…r PowerSGD" Add an index field to GradBucekt, so error_dict is keyed by this index instead of the hashcode of input tensor. Howevever, sometimes the buckets can be rebuilt in the forward pass. In this case, the shape of the bucket with the same index will not be consistent with the one in the previous iteration, and hence the error tensor will be re--initialized as a zero tensor of the new shape. Original PR issue: Investigate Applying PowerSGD to Communication Hook for Gradient Compression #47202 Differential Revision: [D25288496](https://our.internmc.facebook.com/intern/diff/D25288496/) [ghstack-poisoned]
Pull Request resolved: #48757 Add an index field to GradBucekt, so error_dict is keyed by this index instead of the hashcode of input tensor. The replacement will be done in a separate diff, as the definition of this new method somehow couldn't be recognized in the OSS version. Original PR issue: Investigate Applying PowerSGD to Communication Hook for Gradient Compression #47202 ghstack-source-id: 117916685 Differential Revision: [D25288496](https://our.internmc.facebook.com/intern/diff/D25288496/)
…r PowerSGD" Add an index field to GradBucekt, so error_dict is keyed by this index instead of the hashcode of input tensor. Howevever, sometimes the buckets can be rebuilt in the forward pass. In this case, the shape of the bucket with the same index will not be consistent with the one in the previous iteration, and hence the error tensor will be re--initialized as a zero tensor of the new shape. Original PR issue: Investigate Applying PowerSGD to Communication Hook for Gradient Compression #47202 Differential Revision: [D25288496](https://our.internmc.facebook.com/intern/diff/D25288496/) [ghstack-poisoned]
…r PowerSGD" Add an index field to GradBucekt, so error_dict is keyed by this index instead of the hashcode of input tensor. Howevever, sometimes the buckets can be rebuilt in the forward pass. In this case, the shape of the bucket with the same index will not be consistent with the one in the previous iteration, and hence the error tensor will be re--initialized as a zero tensor of the new shape. Original PR issue: Investigate Applying PowerSGD to Communication Hook for Gradient Compression #47202 Differential Revision: [D25288496](https://our.internmc.facebook.com/intern/diff/D25288496/) [ghstack-poisoned]
Pull Request resolved: #48757 Add an index field to GradBucekt, so error_dict is keyed by this index instead of the hashcode of input tensor. The replacement will be done in a separate diff, as the definition of this new method somehow couldn't be recognized in the OSS version. Original PR issue: Investigate Applying PowerSGD to Communication Hook for Gradient Compression #47202 ghstack-source-id: 117939208 Differential Revision: [D25288496](https://our.internmc.facebook.com/intern/diff/D25288496/)
This pull request has been merged in 7439bc4. |
Stack from ghstack:
Add an index field to GradBucekt, so error_dict is keyed by this index instead of the hashcode of input tensor.
Howevever, sometimes the buckets can be rebuilt in the forward pass. In this case, the shape of the bucket with the same index will not be consistent with the one in the previous iteration, and hence the error tensor will be re--initialized as a zero tensor of the new shape.
Original PR issue: Investigate Applying PowerSGD to Communication Hook for Gradient Compression #47202
Differential Revision: D25288496