LoRA — Low-Rank Adaptation
Definition: A fine-tuning technique that adapts a large model by training small adapter layers instead of the full network. Cheap, fast, and swappable — the workhorse of practical model customization.
Example
A LoRA trained on our support tickets matched full fine-tuning quality at 2% of the GPU cost.
When you'll hear it
LoRA shows up most often in AI strategy reviews, model evaluation discussions, and product roadmap meetings. When someone uses it, they're usually referring to low-rank adaptation — and they expect the room to already know what that means.
FAQs
What does LoRA stand for?
LoRA stands for Low-Rank Adaptation.
What does LoRA mean in AI and machine-learning?
A fine-tuning technique that adapts a large model by training small adapter layers instead of the full network. Cheap, fast, and swappable — the workhorse of practical model customization.
Where will I hear LoRA used at work?
LoRA comes up most often in AI strategy reviews, model evaluation discussions, and product roadmap meetings. It's used as shorthand for low-rank adaptation, so people assume you already know the term.