Use case
Customise a model
Adapt a general model to a specific task or style, and keep the cost of doing so proportionate.
What it takes
In the order you would adopt them. Every step opens in the knowledge graph.
-
Family of decoder-only Transformer language models trained with a generative objective.
-
Continue training a pre-trained model on task-specific data to adapt its behaviour.
-
Fine-tuning by training small low-rank adapter matrices instead of all weights.
-
Cached attention keys and values that make generating one token at a time affordable.