fine-tuning a model is very different from training the whole model in terms of resource requirements.