RAG vs LoRA vs Full Fine-Tuning
Same LLM, three very different strategies. Context, adapters, or every weight — know exactly what you're changing before you pick.
01
RAG
Add context · Keep weights
Best for: fresh · private · citable knowledge
02
LoRA Fine-Tuning
Train adapter · Freeze base
Best for: lower-cost task · style · format adaptation
03
Full Fine-Tuning
Train every weight
Best for: deep specialization · enough data + budget
What actually changes?
RAG
Changes the context given to the model. Weights untouched.
LoRA
Changes a small adapter around the model. Base stays frozen.
Full Fine-Tuning
Changes all weights. A new, specialized model.
Don't ask "which one is best?" — ask "what am I trying to change: behavior or knowledge?"
Knowledge gap → RAG · Behavior gap → train