Also known as: Low-Rank Adaptability (LoRA) · Low-Rank Adaptability LoRA · Low Rank Adaptability [LoRA]
Low-Rank Adaptability, обычно называемая LoRA, — параметрически эффективная техника fine-tuning, которая замораживает pretrained model и вставляет небольшой набор обучаемых low-rank матриц обновления. Она обеспечивает более быструю и менее затратную адаптацию к специализированным доменам при сохранении базовой модели. LoRA часто используется для поддержания нескольких task-specific adaptations поверх общей foundation model.
A parameter-efficient technique for fine tuning that "freezes" the model's pre-trained weights (such that they can no longer be modified) and then inserts a small set of trainable weights into the model. This set of trainable weights (also known as "update matrixes") is considerably smaller than the base model and is therefore much faster to train. LoRA provides the following benefits: - Improves the quality of a model's predictions for the domain where the fine tuning is applied. - Fine-tunes faster than techniques that require fine-tuning all of a model's parameters. - Reduces the computational cost of inference by enabling concurrent serving of multiple specialized models sharing the same base model. The update matrixes used in LoRA consist of rank decomposition matrixes, which are derived from the base model to help filter out noise and focus training on the most important features of the model.