Training & post-training
Low-rank adaptation
A parameter-efficient method for adapting a model without updating every weight.
Also known as: LoRA
Definition
Low-rank adaptation trains small additional matrices while keeping most base-model parameters fixed, reducing compute and storage needs.
Why it matters
LoRA enables cheaper domain variants, but every adapter still needs versioning, security review and use-case validation.
Related concepts
- Model parameters
The learned numerical values that determine a model's behaviour.
- Fine-tuning
Additional training that adapts a model to a task, domain or behaviour.
- Model version
A uniquely identifiable release or snapshot of a model.