How to use reg:squaredlogerror or reg:pseudohubererror objective correctly in xgboost for regression?

Question

I am trying to predict y with a range from 0 to 1000000. Using reg:squarederror or reg:linear objective gives an good prediction on the larger valu