Technical

Pre-training

Initial stage where a model learns general patterns from data.

Definition

The stage in model development where a model is trained on large-scale, broad datasets to learn general-purpose representations and patterns prior to task-specific adaptation. Pre-training typically requires substantial compute and data resources and is a key determinant of model capabilities and market concentration.