diff options
| author | Alex Morehead <alex.morehead@gmail.com> | 2024-10-02 02:50:49 -0500 |
|---|---|---|
| committer | GitHub <noreply@github.com> | 2024-10-02 09:50:49 +0200 |
| commit | d661859ad4206217afb274be709837d19218b3f8 (patch) | |
| tree | e4126070c1a6d5e34e4b68a205f9e89b6f2d0b65 /README.md | |
| parent | a3edfae103d79d49ad51aea2cdd0a3dba37d6403 (diff) | |
Diffstat (limited to 'README.md')
| -rw-r--r-- | README.md | 8 |
1 files changed, 4 insertions, 4 deletions
@@ -67,7 +67,7 @@ with ema.average_parameters(): ### Manual validation mode -While the `average_parameters()` context manager is convinient, you can also manually execute the same series of operations: +While the `average_parameters()` context manager is convenient, you can also manually execute the same series of operations: ```python ema.store() ema.copy_to() @@ -77,7 +77,7 @@ ema.restore() ### Custom parameters -By default the methods of `ExponentialMovingAverage` act on the model parameters the object was constructed with, but any compatable iterable of parameters can be passed to any method (such as `store()`, `copy_to()`, `update()`, `restore()`, and `average_parameters()`): +By default the methods of `ExponentialMovingAverage` act on the model parameters the object was constructed with, but any compatible iterable of parameters can be passed to any method (such as `store()`, `copy_to()`, `update()`, `restore()`, and `average_parameters()`): ```python model = torch.nn.Linear(10, 2) model2 = torch.nn.Linear(10, 2) @@ -90,11 +90,11 @@ ema.copy_to(model2) ### Resuming training -Like a PyTorch optimizer, `ExponentialMovingAverage` objects have `state_dict()`/`load_state_dict()` methods to allow pausing, serializing, and restarting training without loosing shadow parameters, stored parameters, or the update count. +Like a PyTorch optimizer, `ExponentialMovingAverage` objects have `state_dict()`/`load_state_dict()` methods to allow pausing, serializing, and restarting training without losing shadow parameters, stored parameters, or the update count. ### GPU/device support `ExponentialMovingAverage` objects have a `.to()` function (like `torch.Tensor`) that can move the object's internal state to a different device or floating-point dtype. -For more details on individual methods, please check the docstrings.
\ No newline at end of file +For more details on individual methods, please check the docstrings. |