Training
Adapter
What is adapter?
An adapter is the small set of trained weights produced by a parameter-efficient fine-tune. It is meaningless on its own and must be applied to the exact base model it was trained against, but it is typically a few megabytes rather than gigabytes.
Small size is what makes several useful things possible: cheap storage of many task variants, near-instant switching between them, and serving many adapters against a single resident copy of the base model.
The pairing is strict. An adapter trained against one version of a base model will produce degraded or nonsensical output against a different version, and nothing in the file format prevents you from trying.
Related terms
Start with the free tier
A magic link creates your account, your tenant and your first API key. No card until you ask for compute.