[1] 9.87e+11

[2] 1.12e+12 which is larger than the initial RF model. A higher learning rate makes each tree contribute more, which can cause overfitting.