Does Spark XGBoost provide any way to free the underlying memory allocated by XGBoost C++ native modules?

Question

We are using Spark to train or retrain models using Spark. For some reasons we want to unload the model from Spark and do some other tasks after training is over. We are not sure how to deallocate all the memory from the XGBoost context as XGBoost is implemented in C++ and Spark is using just the Java wrapper to communicate with XGBoost native APIs.