1. Refactor training_args
2. Refactor hooks
3. Add train_id for push_to_hub
4. Support both output_dir/output_sub_dir for checkpoint_hooks
5. Support copy when hardlink fails when checkpointing
6. Support mixed dataset config file as a CLI argument
7. Add eval txt in output folder
Link: https://code.alibaba-inc.com/Ali-MaaS/MaaS-lib/codereview/12384253
* support the ignorance of file pattern