Given a machine learning model A trained for task 1, one can extract part of A, along with its trained weights, adapt it into model B (say, by adding layers) which is appropriate for task 2, then continue training B on a dataset set up for task 2. In this scenario it is said that the model A has been fine-tuned to solve task 2, or that the model B has been pre-trained on task 1. Fine-tuning is a type of "transfer learning" -- i.e., the adaptation of a model developed for one task to tackle another, related task.