Configurable GradScaler Parameters #1038
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
|
@manuelknott has exported this pull request. If you are a Meta employee, you can view the originating Diff in D86090032. |
manuelknott
force-pushed
the
export-D86090032
branch
from
2dbadfd to
799ac57
Compare
manuelknott
added a commit
to manuelknott/tnt
that referenced
this pull request
Nov 3, 2025
Summary: TorchTNT's `AutoUnit` initializes a GradScaler if `precision == torch.float16` and uses default parameters of `from torch.amp.grad_scaler.GradScaler` or `torch.distributed.fsdp.sharded_grad_scaler.ShardedGradScaler` respectively. Some projects require custom arguments for their GradScaler, thus, I propose to expose these parameters and make them configurable. I propose to implement them as a dataclass analoguous to `SWALRParams` and `SWAParams`. Also added integration tests in `test_precision.py`. Differential Revision: D86090032
Summary: TorchTNT's `AutoUnit` initializes a GradScaler if `precision == torch.float16` and uses default parameters of `from torch.amp.grad_scaler.GradScaler` or `torch.distributed.fsdp.sharded_grad_scaler.ShardedGradScaler` respectively. Some projects require custom arguments for their GradScaler, thus, I propose to expose these parameters and make them configurable. I propose to implement them as a dataclass analoguous to `SWALRParams` and `SWAParams`. Also fixed pre-existing indentation error in callbacks.rst that was blocking build_docs CI (introduced in e6b119b1a422) Differential Revision: D86090032
manuelknott
force-pushed
the
export-D86090032
branch
from
799ac57 to
06768d7
Compare
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
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:
TorchTNT's
AutoUnitinitializes a GradScaler ifprecision == torch.float16and uses default parameters offrom torch.amp.grad_scaler.GradScalerortorch.distributed.fsdp.sharded_grad_scaler.ShardedGradScalerrespectively.Some projects require custom arguments for their GradScaler, thus, I propose to expose these parameters and make them configurable.
I propose to implement them as a dataclass analoguous to
SWALRParamsandSWAParams.Also added integration tests in
test_precision.py.Differential Revision: D86090032