Exponentiate the RMSE, because the RMSE on the log scale corresponds to the geometric mean of the squared errors in the original scale.

Take the square root of the RMSE, because the RMSE on the log scale corresponds to the arithmetic mean of the squared errors in the original scale.

Multiply the RMSE by the mean of the original label variable, because the RMSE on the log scale is proportional to the original scale.

Do not convert the RMSE, because it is already in the original scale of the label variable.