Prompt Tuning
Definition
Adapt learnable prompt vectors to a task.
Prompt tuning trains additional continuous input vectors for a task, while the weights of the underlying model typically remain unchanged.
Simply explained
A small learnable prefix directs an existing model to a task.
Why it matters in practice
Task-specific adaptation can be done with fewer trainable parameters than full fine-tuning.
Example from business
A hypothetical research team trains a soft prompt for a fixed classification task.
Limits and pitfalls
- Learned vectors are not ordinary text instructions and are not arbitrarily transferable between models.
Origin & history
The method belongs to parameter-efficient adaptation of pretrained models.
Use in mid-sized companies
- Consider only with sufficient examples and measurable advantage over simple prompts.
Frequently asked questions
Is this manual rewriting?
No, the additional vectors are trained.
Sources
Editorially reviewed on: