LoRA

LoRA is a technique for fine-tuning large models cheaply: instead of updating billions of weights, it trains small add-on matrices that steer the frozen base model. The result is adapter files small enough to email, which is both the appeal and the risk. Anyone can specialize an open model on a laptop, and a malicious adapter can quietly change a model's behavior, so adapters deserve the same supply-chain scrutiny as any other dependency.