The Power of Scale for Parameter-Efficient Prompt Tuning

Abstract

In this work, we explore "prompt tuning", a simple yet effective mechanism for learning "soft prompts" to condition frozen language models to perform specific downstream tasks. Unlike the discrete text prompts used by GPT-3, soft prompts are learned through backpropagation and can be tuned to incorporate signal from any number of labeled examples. Our end-to-end learned approach outperforms GPT-3's "few-shot" learning by a large margin. More remarkably, through ablations on model size using T5, we show that prompt tuning becomes more competitive with scale: as models exceed billions of parameters, our method "closes the gap" and matches the strong performance of model tuning (where all model weights are tuned). This finding is especially relevant in that large models are costly to share and serve, and the ability to reuse one frozen model for multiple downstream tasks can ease this burden. Our method can be seen as a simplification of the recently proposed "prefix tuning" of Li and Liang (2021), and we provide a comparison to this and other similar approaches. Finally, we show that conditioning a frozen model with soft prompts confers benefits in robustness to domain transfer, as compared to full model tuning.

Learning to Compose Diversified Prompts for Image Emotion Classification

“…While manually crafting prompts [Brown et al, 2020] [Radford et al, 2021is intuitive, creating and experimenting with these prompts takes time and experience, even experienced prompt designers may fail to manually discover optimal prompts . To automate prompt engineering, ] [Lester et al, 2021] [Zhou et al, 2021 paramerized the prompts by treating prompts as virtual tokens and perform prompting directly in the embedding space.…”

“…Following [Lester et al, 2021], we initialize the class-specific prompts p c to maximize the likelihood of P (y pred = y|p c ). However, just like the part (e) in the Figure1, there are significant differences among the content of different affective images even though they are in the same class.…”

“…(3) Shared invariant prompts: Some work in NLP treat prompts as sequences of virtual tokens and learns prompts automatically by parameterizing these tokens [Lester et al, 2021], avoiding the suboptimal problem to some extent . However, these methods employed shared prompts across all instances, regardless of the fact that instances of different categories share similar features while also have their own distinct characteristics.…”

A Survey of Controllable Text Generation using Transformer-based Pre-trained Language Models

“…Different from the traditional approaches that encode the sentence into a set of vectors and then classify their sentiment through a fully connected layer, the prompt-based method will construct a set of templates, for example: ("I am always happy to see you, the sentence's sentiment is [MASK]"), and then ask the model to predict the token [mask] according to the original training task for the PLM. This approach has gone through various stages, from manual template construction [Jiang et al 2020], to automated search for discrete tokens [Shin et al 2020], to continuous virtual Tokon representations [Lester et al 2021;Li and Liang 2021]. It has achieved a great success in few-shot scenarios.…”

“…The approach achieves impressive results on some generative tasks such as data-to-text. An extension of the model, namely P-tuning [Lester et al 2021], serves a similar purpose. Different from prefix-tuning [Li and Liang 2021], p-tuning does not place prompt with the "prefix" in the input, but constructs a suitable template to prompt the PLM, and the template is composed of continuous virtual token which is obtained through gradient descent.…”