Le prompt tuning est une méthode d’adaptation efficace en paramètres qui apprend un petit ensemble de vecteurs propres à une tâche, souvent appelé soft prompt ou préfixe, tout en laissant la plupart ou la totalité des paramètres du modèle de base gelés. Le prompt appris est ajouté au début des entrées du modèle afin d’améliorer les performances sur une tâche définie à un coût inférieur au fine-tuning complet.
A parameter efficient tuning mechanism that learns a "prefix" that the system prepends to the actual prompt. One variation of prompt tuning---sometimes called prefix tuning ---is to prepend the prefix at every layer . In contrast, most prompt tuning only adds a prefix to the input layer. For prompt tuning, the "prefix" (also known as a "soft prompt") is a handful of learned, task-specific vectors prepended to the text token embeddings from the actual prompt. The system learns the soft prompt by freezing all other model parameters and fine-tuning on a specific task.
Prompt tuning is a parameter-efficient fine-tuning method used to improve a generative AI model's performance on a specific task. It involves learning a "prefix" that is prepended to the actual prompt, sometimes at every layer. This approach is considered cheaper and faster than other tuning methods, often yielding good results. Prompt tuning is particularly effective when you have a specific task and want the model to perform it in a certain way. It's also sometimes referred to as prompt learning or parameter-efficient (fine) tuning. For more information, see Introduction to prompt engineering.
A side-by-side comparison of Prompt Engineering and Prompt Tuning. Understand how human-designed model inputs differ from learned task-specific soft prompts.
A side-by-side comparison of Prompt Tuning and Fine-tuning. Understand how a learned soft prompt differs from retraining a pretrained model on a specialized dataset.