Definition
LoRA / QLoRA
Parameter-efficient fine-tuning methods that train small adapter matrices instead of the whole model, cutting compute and memory needs.
LoRA (Low-Rank Adaptation) freezes the original model and trains small additional matrices that adjust its behavior. QLoRA combines this with quantization so large open-source models can be fine-tuned on modest GPUs.
Adapters are small files that can be swapped per task or customer, which makes LoRA practical for teams running open-source models such as Llama or Mistral in their own infrastructure.
Related services
Related terms