Concept
How LLMs are trained
The pipeline from data and tokenization through pre-training to fine-tuning and preference tuning.
What should you learn before How LLMs are trained?
- Deep learningMachine learning with many-layered neural networks.
- Tokens and tokenizationSplitting text into the units a language model reads and is billed by.
- TransformersThe attention-based neural network architecture behind modern language models.
- Large language models (LLMs)Transformer models trained on large amounts of text to predict the next token.
What to learn after How LLMs are trained
- Fine-tuningFine-tuning continues training a pre-trained model.
Connections to other concepts
- Includes
- AI safety and alignment