AI Glossary / Model training
🏗️ Models and training
Model training
The process of "teaching" a network on huge amounts of data - this is when the model acquires its capabilities.
Training is the process where a model learns its capabilities: a neural network processes huge amounts of data, and its parameters are gradually tuned to increasingly better predict correct answers. For language models, this means weeks or months of computation on thousands of graphics cards. After training is complete, the model's knowledge is "frozen" - which is why models have a knowledge cutoff date and don't know about more recent events, unless connected to search. Training from scratch is expensive, so fine-tuning existing models is more common.
Related concepts