Fit the model using entire data or from training data?

Question

I am given two data.

Firstly, the train data with known class (target)

Secondly, the test data with no class (no target)

I split the training data into train set and validation set . I oversample the train data and test it on my validation set.

It is an imbalanced dataset.

After picking out the best model, Will I fit it back to my entire dataset for my final prediction on test(unseen data)

Model = LGBMClassifier()

Model.fit(X,Y)

Model.predict (test)

or I fit it on oversample training .

Model = LGBMClassifier()

Model.fit(X_train_smote,Y_train_smote)

Model.predict (test)