LoRA (Low-Rank Adaptation)
Also known as: Low-Rank Adaptation, PEFT
Fine-tune by training small low-rank update matrices instead of every weight: same quality, ~1% of the cost.
A parameter-efficient fine-tuning method that injects trainable low-rank matrices into frozen model layers. Reduces GPU memory and training cost by 10-100x while achieving near-full fine-tuning quality.
In practice
The default fine-tuning approach for cost-conscious teams. Expect questions on rank choice, target modules, and merging adapters at inference.
How it compares
LoRA is a parameter-efficient variant of fine-tuning; full FT updates every weight.
Comparisons that include LoRA (Low-Rank Adaptation)
Related topics
Related terms
Fine-tuning
Continue training a pre-trained model on task-specific data to specialize its behavior.
DPO (Direct Preference Optimization)
Skip the reward model and PPO: fine-tune directly on preferred-vs-rejected response pairs.
Knowledge Distillation
Train a small student model to match a big teacher's outputs: cheap, fast inference with most of the quality.
Pre-training
Self-supervised next-token training on a huge unlabeled corpus, producing the base model.
Supervised Fine-Tuning (SFT)
Fine-tune the base model on (prompt, ideal response) pairs; the first post-training step before alignment.
QLoRA
LoRA combined with 4-bit base-model quantization, letting you fine-tune huge models on a single consumer GPU.