Glossary · Training

Fine-tuning

Also known as: FT, Continued training

Continue training a pre-trained model on task-specific data to specialize its behavior.

Adapting a pre-trained model to a specific task or domain by continuing training on a smaller, curated dataset. Parameter-efficient methods like LoRA modify only a fraction of the weights, reducing cost and catastrophic forgetting.

In practice

The classic "do we fine-tune or RAG?" question shows up in nearly every applied LLM system-design round.

How it compares

vs. Retrieval-Augmented Generation (RAG) →Deep comparison →

Fine-tuning changes weights and is best for new behavior; RAG keeps weights frozen and is best for new knowledge.

vs. LoRA (Low-Rank Adaptation) →Deep comparison →

LoRA is a parameter-efficient way to fine-tune; full fine-tuning updates every weight.

Comparisons that include Fine-tuning

In-context learning vs Fine-tuningIn-context learning adapts at inference by putting examples in the prompt; fine-tuning adapts by updating weights. Reach for prompting first; fine-tune when you cannot get there.LoRA vs Full Fine-tuningLoRA reaches 90-99% of full fine-tuning quality for a fraction of the compute and storage; use full fine-tuning only when that last margin is genuinely worth it.RAG vs Fine-tuningReach for RAG when knowledge changes or must be cited; fine-tune when you need the model to change how it behaves, not what it knows.

Questions that mention this term

Related terms

LoRA (Low-Rank Adaptation)

Fine-tune by training small low-rank update matrices instead of every weight: same quality, ~1% of the cost.

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.