Customizing LLMs: Fine-tuning, LoRA & RAG
Make a model your own — without guesswork. Learn when to fine-tune, when to reach for RAG, and how LoRA, distillation and good evaluation actually work, remembered with spaced repetition.
- flashcards
- 74
- flashcards
- per day
- ~10 min
- per day
- level
- Intermediate
- level
- modules
- 5
- modules
What does customizing a model mean?
Out of the box, a general-purpose LLM knows a lot but nothing about your domain, your data or your house style. Customizing a model is how you close that gap — and the first real skill is knowing which tool the job needs, because the options trade off cost, freshness and effort very differently.
Broadly there are two levers. Fine-tuning adjusts the model's own weights so it internalizes a behavior or format; RAG leaves the weights alone and feeds relevant facts in at query time. Around them sit the efficiency techniques — LoRA and PEFT to fine-tune cheaply, distillation and quantization to shrink and speed up — plus the data and evaluation work that decides whether any of it actually helped.
This track breaks customization into bite-sized, practical questions and uses spaced repetition so the trade-offs stick — so you reach for fine-tuning, RAG or a smaller quantized model on purpose, not by habit.
5 modules, seed to bloom
Each module is a set of flashcards — 74 in total. Answer, review, and watch your knowledge grow from seed to full bloom.
Fine-tuning Fundamentals
Fine-tuning fundamentals — what it is, what it teaches, when it helps versus prompting or RAG, and its risks.
15 cardsLoRA & PEFT
LoRA and PEFT — parameter-efficient fine-tuning, why it is cheap, adapters, QLoRA, and rank trade-offs.
15 cardsRAG vs Fine-tuning
RAG versus fine-tuning — which to choose for facts, behavior, freshness, citation, and cost, and how to combine them.
14 cardsDistillation & Quantization
Distillation and quantization — shrinking and speeding up models, precision trade-offs, and how the techniques differ.
15 cardsData & Evaluation
Data and evaluation — what data fine-tuning needs, validation sets, measuring improvement, and avoiding overfitting.
15 cardsSample questions
A taste of the real flashcards. Pick an answer, then reveal the explanation.
What is fine-tuning of an LLM?
- AFurther training a pretrained model on custom data to adjust its weights
- BWriting detailed prompts that steer a pretrained model at request time
- CAttaching a document store so the model can look facts up at query time
- DCompressing a pretrained model to lower precision for cheaper inference
What is PEFT (parameter-efficient fine-tuning)?
- AFine-tuning only a small subset of parameters instead of all of them
- BFine-tuning every weight, but for far fewer steps than a full run
- CFine-tuning one layer at a time, to keep the optimizer state in memory
- DFine-tuning on data the model generated itself, to avoid labelling cost
What is the best approach to give a model up-to-date facts?
- ARAG — retrieve current documents and add them to the prompt
- BFine-tuning — bake the facts into the model's weights once
- CDistillation — compress the facts into a smaller student model
- DQuantization — lower the precision so more facts fit in memory
What is knowledge distillation?
- ATraining a small student model to imitate a larger teacher model
- BTraining a large model on the outputs of many smaller models
- CRemoving low-importance weights to shrink an already-trained model
- DLowering the precision of a model's weights to save on memory
Learn it once, keep it for good
Answer a question
Each card is one practical concept with multiple options. Pick what you think is right.
Get the full answer
See the correct option plus a clear explanation, and a link to deeper docs when one is available.
Review at the right time
A spaced-repetition engine (SM-2 or FSRS) resurfaces each card just before you would forget it.
Why customization is worth your time
Pick the right tool
Fine-tune, RAG or prompt? Knowing the trade-offs saves weeks of effort spent on the wrong approach.
Fine-tune on a budget
LoRA, QLoRA and PEFT make adapting a model affordable — understanding them is the difference between feasible and not.
Ship smaller, faster models
Distillation and quantization cut size and latency; knowing the precision trade-offs keeps quality where you need it.
Know if it worked
Validation sets and honest evaluation are what separate a real improvement from a confident-looking regression.
Common questions
Should I understand how LLMs work first? +
It helps a lot. This track assumes you know the basics of tokens, training and prompting — the AI & LLM Fundamentals and Prompt Engineering tracks pair naturally with it.
Fine-tuning or RAG — which should I learn? +
Both, because the real skill is choosing between them. A whole module is dedicated to RAG vs fine-tuning: facts and freshness usually favor RAG, behavior and format favor fine-tuning, and often you combine them.
Is it free? +
Yes, completely free. No registration or credit card is required, and all your progress is stored locally in your browser.
How long does it take? +
About 10 minutes a day. Spaced repetition means short, frequent sessions beat long cramming, so the trade-offs stick.
Ready to customize models with intent?
Plant your first seed today. Ten minutes a day is all it takes to know exactly when to fine-tune, when to RAG, and when to leave the model alone.