Why is F1 score of NER lower after optimizng CRF hyperparameters on development set?

Question

from sklearn_crfsuite import CRF, metrics

X_train,y_train
X_dev,  y_dev
X_test, y_test

# baseline run
baseline = CRF(algorit