Models & architecture
Foundation model
A large pre-trained model adaptable to many downstream tasks.
Definition
A foundation model is trained broadly at scale and can be prompted, fine-tuned or connected to tools and data for multiple applications.
Why it matters
A bank may reuse one foundation model across many systems, creating concentration, inheritance and change-management risks.
Related concepts
- Large language model
A foundation model trained to understand and generate sequences of language tokens.
- Pretraining
Initial large-scale training that gives a foundation model broad capability.
- System under test
The complete configuration being evaluated—not only the model.