XGBoost made an improvement to support large model size in PR https://github.com/dmlc/xgboost/pull/10984, I just create an issue to track if spark-rapids-ml needs it.