Understanding Parameter-Efficient LLM Finetuning: Prompt Tuning And Prefix Tuning

Last week, I shared an overview of different Large Language Model (LLM) finetuning techniques. In a series of short articles, I am planning to discuss a selection of the most relevant techniques one by one.

Finetuning Large Language Models

Let's start with a selection of parameter-efficient finetuning techniques concerning prompt modifications.

A selection of parameter-efficient finetuning techniques covered in this article.

Prompt Tuning

The original concept of prompt tuning refers to techniques that vary the input prompt to achieve better modeling results. For example, suppose we are interested in translating an English sentence into German. We can ask the model in various different ways, as illustrated below.

An example of hard prompt tuning, that is, rearranging the input to get better outputs.

Now, this concept illustrated above is referred to as hard prompt tuning since we directly change the discrete input tokens, which are not differentiable.