v0.3.0

[0.3.0] - 2021-11-17

Added

• torch_ema.__version__

Changed

• Parameters without requires_grad = True are no longer partially ignored, resolving #9; now all parameters passed to the EMA object have EMA run on them, regardless of whether they are trainable or not.

v0.2

Version 0.2 of pytorch_ema. If you upgrade and have new backward compatibility problems, this release is probably what you need as a workaround.