The initial training phase where a model learns general patterns from a large dataset before being adapted for specific downstream tasks.