Further training of an already-trained model on new data to extend or improve its capabilities without starting from scratch.
Performance retention over long documents and conversations