- Added `evaluate_during_training_steps` parameter that specifies when evaluation should be performed during training.
Changed
- A model checkpoint will be created for each evaluation during training and the evaluation results will be saved along with the model.
0.15.0
0.14.00.14.0
Added
- Added option to specify a GPU to be used when multiple GPUs are available. E.g.: `cuda_device=1` - Added `do_lower_case` argument for uncased models.
Fixed
- Fixed possible bug with output directory not being created before evaluation is run when using `evaluate_during_training` .