LoRA (Low-Rank Adaptation)

A parameter-efficient fine-tuning technique that adds small trainable matrices to pre-trained models, reducing the computational cost of customization by up to 10,000x. LoRA enables organizations to fine-tune large models on consumer hardware, making custom AI accessible to startups and researchers.