XGBRegressor conversion to onnx gives wrong predictions for xgboost 2.0

[veselina-marinova-cresset]

Following the procedure for conversion to onnx outlines here: https://onnx.ai/sklearn-onnx/auto_tutorial/plot_gexternal_xgboost.html , I found that when trying to load a XGBoost regression model the predictions are wrong. The problem is present for xgboost 2.0, but not for xgboost 1.7