Back to Home
Training Fundamentals
🟢 Beginner · 7–9 min

Pretraining

Pretraining is the first major learning stage. The model studies a large amount of text and learns general patterns in language, long before it is given any specific job.

Pretraining Playground

Feed the data, build a base model

A model starts out knowing nothing. Press Start Pretraining and watch large text data flow into the model. As it reads, it slowly picks up patterns and becomes a base model.

Large Text Data
Pretraining
The Model
Empty, knows nothing yet
Result
Base Model
Not ready yet
Ready, press Start Pretraining0%
Patterns the model picks up
The model learns patterns, not exact understanding like a human.
What the Model Learns

Click a pattern to see an example

Pretraining is not about memorising facts. The model soaks up broad, general patterns. Tap each card below to see what that looks like in practice.

Base Model vs Specialized Model

Two stages, two kinds of model

Pretraining creates the base model. Fine-tuning later makes it more useful for specific tasks.

Base Model
Created after pretraining.
Knows many general patterns
Understands broad language and facts
Not yet tuned for assistant behavior
Specialized Model
Created after fine-tuning.
Built on top of the base model
Better at following specific tasks
Tuned to follow instructions well
🔑
Key takeaways
  • • Pretraining is the first major training stage.
  • • The model learns from large amounts of data.
  • • It learns general language and knowledge patterns.
  • • Pretraining creates a base model.
  • • Fine-tuning can later specialize the base model.
Summary

Pretraining gives an AI model its general foundation. It learns broad patterns from large data before being adapted for more specific tasks.