accelerate

27a142bc - fix(tracking): default step=None on tracker.log and accept extra kwargs in MLflowTracker (#4039)

fix(tracking): default step=None on tracker.log and accept extra kwargs in MLflowTracker (#4039) * fix(tracking): make `step` optional and accept **kwargs in MLflowTracker.log Several tracker `log` / `log_images` methods declared `step: Optional[int]` without a default, even though their docstrings said the parameter was optional. Calling `tracker.log(values)` therefore raised `TypeError: log() missing 1 required positional argument: 'step'` on `GeneralTracker`, `TensorBoardTracker.log_images`, `AimTracker.log`, and `MLflowTracker.log`. `MLflowTracker.log` also did not accept `**kwargs`, so passing per-tracker arguments via `Accelerator.log(log_kwargs={"mlflow": {...}})` raised `TypeError: log() got an unexpected keyword argument`. The forwarded kwargs are now passed through to `mlflow.log_metrics`. Added two regression tests on `MLflowTracker` covering both paths. * test: assert MLflowTracker.log forwards extra kwargs to log_metrics