[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.