Prepare GradScaler for hivemind.Optimizer #413
Merged
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
Member
justheuristic
commented
Nov 18, 2021
•
justheuristic
commented
- Modified hivemind.GradScaler to make it compatible with hivemind.Optimizer (backwards-compatible)
- Changed TrainingStateAverager to be compatible with hivemind.GradScaler
- Made TrainingStateAverager.main_parameters and parameter_names public for use in optimizer
justheuristic
commented
Nov 18, 2021
| @@ -100,7 +100,7 @@ def __init__( | |||
| self.offload_optimizer = offload_optimizer | |||
| self.custom_gradients = custom_gradients | |||
|
|
|||
| self._main_parameters, self._parameter_names = main_parameters, parameter_names | |||
| self.main_parameters, self.parameter_names = main_parameters, parameter_names | |||
There was a problem hiding this comment.
made them public because these params are needed in hivemind.Optimizer.step
... and they are no more private than, for instance, opt_keys_for_averaging
mryab
reviewed
Nov 18, 2021
| @@ -378,20 +378,31 @@ def step( | |||
| self.finished_optimizer_step.clear() | |||
| return output | |||
|
|
|||
| def _do(self, optimizer_step: bool, zero_grad: bool, averaging_round: bool, **kwargs): | |||
| def _do(self, optimizer_step: bool, zero_grad: bool, averaging_round: bool, grad_scaler: Optional[GradScaler], **kwargs): | |||
There was a problem hiding this comment.
Let's discuss this with @borzunov on the main PR
hivemind/optim/grad_scaler.py
Outdated
| def unscale_(self, optimizer: Optimizer) -> bool: | ||
| assert isinstance(optimizer, DecentralizedOptimizerBase) | ||
| def unscale_(self, optimizer: TorchOptimizer) -> bool: | ||
| assert hasattr(optimizer, "opt"), "hivemind.GradScaler only supports hivemind optimizer wrappers" |
There was a problem hiding this comment.
What if there's a non-hivemind wrapper of TorchOptimizer? A more explicit check would use isinstance IMO
mryab
approved these changes
Nov 18, 2021
mryab
changed the title
Codecov Report
@@ Coverage Diff @@
## master #413 +/- ##
==========================================
- Coverage 84.61% 84.27% -0.35%
==========================================
Files 76 76
Lines 7286 7287 +1
==========================================
- Hits 6165 6141 -24
- Misses 1121 1146 +25
|
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.