CHANGES PARAMETERS
Fine-tuning
Additional training on selected examples.
↓ Instruction Tuning
↓ Preference Learning / RLHF
04 · Training & Model Adaptation
How Models Learn — and How We Adapt Them
This module explains how models learn from training data, how broad pre-training creates a base model, and how different adaptation methods change behavior or provide new context.
Understand what changes model parameters and what only changes runtime context.
The Big Picture · training and adaptation fork
TRAINING DATA ↓ PRE-TRAINING ↓ FOUNDATION MODEL ↓ HOW DO WE ADAPT IT?
RUNTIME CONTEXT
Prompting
Changes instructions at runtime.
RAG
Adds retrieved external information to context.
Knowledge Structure
A · LEARN FROM DATA
Training Data
The examples used during training.
Synthetic Data
Artificially generated examples for training, evaluation, and testing.
Pre-training
Broad initial training for general capabilities.
B · CHANGE THE MODEL
Fine-tuning
Additional training on selected examples.
Instruction Tuning
Instruction-response training.
Preference Learning / RLHF
Preference feedback shapes behavior.
C · CHANGE REQUEST ENVIRONMENT
Prompting
Changes instructions at runtime.
RAG
Adds external information at runtime.
Key Distinctions
Pre-trainingBroad initial training≠Fine-tuningTargeted additional training
Fine-tuningChanges learned behavior≠PromptingChanges runtime instructions
RAGRuntime context≠Fine-tuningParameter-changing training
Recommended Learning Path
This is a learning path, not one required training sequence.
01Training Data
What models learn from.
02Synthetic Data
Generated examples for controlled training and testing.
03Pre-training
Create general capabilities.
04Fine-tuning
Change behavior through training.
05Instruction Tuning
Train on instructions.
06Preference Learning
Use feedback.
07Prompting vs RAG vs Fine-tuning
Choose an adaptation route.
Remember the distinction
Training changes what the model has learned; prompting and RAG usually change what the model receives when it is used.