What does LoRA stand for?

• A. Learned Reweighting and Adaptation.
• B. Lightweight Recurrent Augmentation.
• C. Latent Reserve Activation.
• D. Low-Rank Adaptation.

Answer: (D)

LoRA stands for Low-Rank Adaptation. The idea is to adapt a large pre-trained model by injecting trainable, low-rank matrices into its layers, while keeping the original weights largely fixed. This lets the model learn task-specific adjustments within a compact subspace, giving parameter-efficient fine-tuning: far fewer new parameters to train and less memory usage, with the ability to preserve the base model’s capabilities. The other options don’t describe this approach or the term used in practice, as they don’t reflect the concept of using low-rank updates to adapt a model.