New features:
- support multi-gpu prediction and evaluation
- verbose argument in system's methods
- CI/CD with travis
Breaking Changes:
- SequenceLabelingGenericPointWiseLossWrapper renamed to TokenLabelingGenericPointWiseLossWrapper
- SequenceLabelingEvaluatorWrapper renamed to TokenLabelingEvaluatorWrapper
Bug Fixes:
- system's train_on_multi_gpus returns results history like train
New Examples:
- BERT for text classification
- ordered sampling for faster training