Training data is the information a model learns from during training. The examples, text, images, labels, feedback, or other data used during training shape what patterns the model can learn.
What is it?
Training data provides examples. Training changes the model.
01Training dataExamples the model can learn from.
02Training processThe model compares predictions with examples.
03Parameters changeInternal values adjust to capture patterns.
04Trained modelThe model behaves differently on new inputs.
Think of it like...
Learning from many examples.
A student studies books, examples, practice questions, and feedback. Over time, the student develops knowledge and patterns.
Study material→ Training data
Learning process→ Training
What the student learns→ Model parameters and behavior
This analogy is for understanding only. Real model training is mathematical and much more complex.
Types of training data
TextWords and symbolsBooks · websites · documents · code
ImagesVisual patternsPhotos · diagrams · visual data
Audio / videoTime-based signalsSpeech · sound · moving visual data
Labels / feedbackSignals about qualityAnswers · categories · preferences · human feedback
Different model types use different combinations of training data.
One simple example · Email classification
Training example 01
Email “Win a free prize”
Label Spam
Training example 02
Email “Meeting at 3 PM”
Label Not Spam
EXAMPLESMany labelled emailsWords and labels reveal useful patterns.
MODEL LEARNS PATTERNS
NEW EMAIL“Claim your prize”A new message arrives.
PREDICTIONSpamSupervised learning connects examples to labels.
Why data quality matters
QualityUseful examplesClear, accurate signals.
CoverageEnough varietyImportant cases are represented.
BalanceFair representationOne group or pattern does not dominate.
Noise / biasLimits and distortionsProblems can affect what the model learns.
Poor, incomplete, biased, or low-quality data can affect model behavior. Better data does not automatically guarantee a perfect model.
Key distinctions
Training Data vs Prompt
Training data: used during training.
Prompt: supplied when using a trained model.
Training data helps shape model parameters; prompts guide a current run.
Training Data vs Context
Training data: used to learn.
Context: information available during inference.
Training Data vs Fine-tuning Data
Training data: the broad category.
Fine-tuning data: data used specifically for additional model adaptation.