In the previous chapters, we explored the basic scenarios of using pre-trained models. A pre-trained model means that it has already been trained to perform specific tasks. Each model is trained on a particular dataset. Two models with the same architecture may behave completely differently if they were trained on different datasets. No matter how large the number of parameters in a model is or how large the dataset it was trained on, there will always be a domain or area of knowledge that was not covered by the dataset used to train the original model. This raises a valid question: what if we need a model that possesses specific knowledge that was not originally embedded in it? And here we arrive at the solution to this problem, known as model training.