No, that should not be fine because this way you will expose your entire dataset (both the train and test dataset). This will lead to bias in the model as the model will be already aware of the test dataset.