search

thunlp / OpenPrompt

An Open-Source Framework for Prompt-Learning.
https://thunlp.github.io/OpenPrompt/
Apache License 2.0
4.32k stars 447 forks source link

Some questions about prompt tuning #267

Closed shaoyuyoung closed 1 year ago

shaoyuyoung commented 1 year ago

hello,

Thanks for your useful work.

I have some questions about prompt tuning.

In #135, one of contributors said that soft prompt + freeze model = prompt tuning.

But I want to know if I just use soft prompt and set freeze model==False. Is it also should be called prompt-tuning?

Thanks in advance

ningding97 commented 1 year ago

This is a subjective question, and I think we should not be entangled with this issue. Technically, the original paper on prompt tuning particularly emphasizes that the model should be frozen. This is more like a "parameter-efficient tuning" method rather than a "prompt-learning" method.

shaoyuyoung commented 1 year ago

Thanks a lot