Caesar AI Atlas

Предобученная модель

Also known as: Pre-Trained Models · Pre trained Model

Caesar AI Atlas Definition

Предобученная модель — это ИИ-модель, обученная на широком или крупном наборе данных до адаптации для конкретной задачи. Она может предоставлять повторно используемые общие возможности, которые затем уточняются с помощью fine-tuning, prompting или других методов кастомизации.

Other Definitions

Предобученная модель Source

Although this term could refer to any trained model or trained embedding vector, pre-trained model now typically refers to a trained large language model or other form of trained generative AI model. See also base model and foundation model.

Предобученная модель Source

‘a component of the training stage in which a model learns general patterns, features, and relationships from vast amounts of unlabeled data, such as through self-supervised learning. Pre-training can equip models with knowledge of general features or patterns which may be useful in downstream tasks, and can be followed with additional training or fine-tuning that specializes the model for a specific downstream task.’

Also Referenced In

Related Terms