How to use adapters
Adapters are the downloadable outputs from a completed run. This page explains what they are, where to use them in your model workflow, and how to test them with fewer surprises.
What an adapter is
An adapter is a lightweight model artifact trained on your dataset. You apply it to a compatible base model to steer outputs toward your task without replacing your full stack.
Where it is used
- In your existing model stack where compatible adapters are supported.
- In evaluation and staging before any production rollout.
- Alongside your prompt and safety controls.
Common outcomes
- Improves behavior on examples similar to your training data.
- Helps consistency for your target task and style.
- Does not guarantee perfect outputs on every edge case.
Before you test it
- Confirm your base model is compatible with adapter usage.
- Prepare a representative test set and success criteria.
- Compare baseline outputs vs adapter outputs side by side.
- Review limits and retention so you download artifacts in time.