What Is Prompt Tuning? Learning a Prompt Instead of Writing One
Instead of hand-crafting a prompt, prompt tuning learns a set of continuous input values automatically, optimized specifically to steer the model toward a target task.
Learning the prompt automatically instead of writing it
Prompt tuning learns a small set of continuous, numeric input values — not actual words a human wrote — that get prepended to a prompt and optimized through training specifically to steer a frozen model toward better performance on a target task, rather than relying on hand-crafted, human-written prompt wording.
Why "learned" input differs from an ordinary prompt
Because these learned values aren't constrained to be actual, readable words, they can encode a more precise, optimized signal for steering the model than any human-written instruction could — at the cost of the result being uninterpretable, since it isn't natural language a person can read and understand directly.
Where it fits among fine-tuning approaches
Prompt tuning is one of the lightest-weight members of the parameter-efficient fine-tuning family — it doesn't touch the model's internal weights at all, only the input, making it cheaper than LoRA or adapter layers, though generally less powerful for harder adaptation tasks.
Frequently Asked Questions
What is prompt tuning?
A technique that learns a small set of continuous input values, optimized through training to steer a frozen model toward a target task, rather than relying on hand-crafted, human-written prompt wording.
Is prompt tuning the same as writing a better prompt manually?
No — the learned values aren't actual readable words a human wrote, but optimized numeric inputs that can encode a more precise steering signal than manual prompt engineering, though the result isn't human-interpretable.