Fine-tuning

Definition

Fine-tuning is the further training of an already pre-trained model to a specific data set or task. Instead of starting from scratch, a basic model that has already learned general patterns (e.g. a language model) is used and adapted to a special area with relatively little data.

Procedure

• Load the base model (e.g. GPT, BERT).

• Provide additional data with labels or desired target task.

• Training with a significantly lower learning rate and often shorter training time until the model delivers accurate results.

Benefits

Fine-tuning saves enormous resources (time, computing power) as the entire model does not have to be relearned. In addition, higher accuracy is often achieved because the model already "fundamentally understands" how speech/images etc. work and only details need to be added for the new task.