Support mixed and low precision training #764
Open
Conversation
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
|
This pull request was exported from Phabricator. Differential Revision: D72415906 |
iden-kalemaj
added a commit
to iden-kalemaj/opacus
that referenced
this pull request
May 30, 2025
Summary: We add support for mixed and low precision training in Opacus. Mixed precision training is supported iwth "hooks", "ghost" grad_sample_modes. Low-precision trainig is additionally supported with "functorch". Why there is no functorch support for mixed precision trainig: The backward pass with functorch performs both a forward and backward pass to compute per-sample gradients. The forrwad pass happens outside of the torch.amp context, so it cannot handle mixed precision. Support for low and mixed precision training is GPU dependent. Differential Revision: D72415906
f8f32d2 to
99b5d14
Compare
|
This pull request was exported from Phabricator. Differential Revision: D72415906 |
iden-kalemaj
added a commit
to iden-kalemaj/opacus
that referenced
this pull request
May 30, 2025
Summary: Pull Request resolved: pytorch#764 We add support for mixed and low precision training in Opacus. Mixed precision training is supported iwth "hooks", "ghost" grad_sample_modes. Low-precision trainig is additionally supported with "functorch". Why there is no functorch support for mixed precision trainig: The backward pass with functorch performs both a forward and backward pass to compute per-sample gradients. The forrwad pass happens outside of the torch.amp context, so it cannot handle mixed precision. Support for low and mixed precision training is GPU dependent. Differential Revision: D72415906
99b5d14 to
11fcba8
Compare
iden-kalemaj
added a commit
to iden-kalemaj/opacus
that referenced
this pull request
Jun 1, 2025
Summary: We add support for mixed and low precision training in Opacus. Mixed precision training is supported iwth "hooks", "ghost" grad_sample_modes. Low-precision trainig is additionally supported with "functorch", "ew" Why there is no functorch support for mixed precision trainig: The backward pass with functorch performs both a forward and backward pass to compute per-sample gradients. The forrwad pass happens outside of the torch.amp context, so it cannot handle mixed precision. Support for low and mixed precision training is GPU dependent. Differential Revision: D72415906
11fcba8 to
3e74b9e
Compare
|
This pull request was exported from Phabricator. Differential Revision: D72415906 |
iden-kalemaj
added a commit
to iden-kalemaj/opacus
that referenced
this pull request
Jun 2, 2025
Summary: We add support for mixed and low precision training in Opacus. Mixed precision training is supported iwth "hooks", "ghost" grad_sample_modes. Low-precision trainig is additionally supported with "functorch", "ew" Why there is no functorch support for mixed precision trainig: The backward pass with functorch performs both a forward and backward pass to compute per-sample gradients. The forrwad pass happens outside of the torch.amp context, so it cannot handle mixed precision. Support for low and mixed precision training is GPU dependent. Differential Revision: D72415906
3e74b9e to
f2b44aa
Compare
|
This pull request was exported from Phabricator. Differential Revision: D72415906 |
Open
iden-kalemaj
added a commit
to iden-kalemaj/opacus
that referenced
this pull request
Jun 2, 2025
Summary: We add support for mixed and low precision training in Opacus. Mixed precision training is supported with "hooks", "ghost" grad_sample_modes. Low-precision training is additionally supported with "functorch", "ew" Why there is no functorch support for mixed precision trainig: The backward pass with functorch performs both a forward and backward pass to compute per-sample gradients. The forrwad pass happens outside of the torch.amp context, so it cannot handle mixed precision. Support for low and mixed precision training is GPU dependent. Differential Revision: D72415906
f2b44aa to
e2fb3db
Compare
|
This pull request was exported from Phabricator. Differential Revision: D72415906 |
Summary: We add support for mixed and low precision training in Opacus. Mixed precision training is supported with "hooks", "ghost", "functorch" grad_sample_modes. Low-precision training is additionally supported with "ew" Support for low and mixed precision training is GPU dependent. Differential Revision: D72415906
e2fb3db to
825defb
Compare
|
This pull request was exported from Phabricator. Differential Revision: D72415906 |
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.
Summary:
We add support for mixed and low precision training in Opacus.
Mixed precision training is supported iw 8000 th "hooks", "ghost" grad_sample_modes.
Low-precision trainig is additionally supported with "functorch".
Why there is no functorch support for mixed precision trainig: The backward pass with functorch performs both a forward and backward pass to compute per-sample gradients. The forrwad pass happens outside of the torch.amp context, so it cannot handle mixed precision.
Support for low and mixed precision training is GPU dependent.
Differential Revision: D72415906