Saved in:
Bibliographic Details
Main Authors: Hu, Jerry Yao-Chieh, Wang, Wei-Po, Gilani, Ammar, Li, Chenyang, Song, Zhao, Liu, Han
Format: Preprint
Published: 2024
Subjects:
Online Access:https://arxiv.org/abs/2411.16525
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915329660157952
author Hu, Jerry Yao-Chieh
Wang, Wei-Po
Gilani, Ammar
Li, Chenyang
Song, Zhao
Liu, Han
author_facet Hu, Jerry Yao-Chieh
Wang, Wei-Po
Gilani, Ammar
Li, Chenyang
Song, Zhao
Liu, Han
contents We investigate the statistical and computational limits of prompt tuning for transformer-based foundation models. Our key contributions are prompt tuning on \emph{single-head} transformers with only a \emph{single} self-attention layer: (i) is universal, and (ii) supports efficient (even almost-linear time) algorithms under the Strong Exponential Time Hypothesis (SETH). Statistically, we prove that prompt tuning on such simplest possible transformers are universal approximators for sequence-to-sequence Lipschitz functions. In addition, we provide an exponential-in-$dL$ and -in-$(1/ε)$ lower bound on the required soft-prompt tokens for prompt tuning to memorize any dataset with 1-layer, 1-head transformers. Computationally, we identify a phase transition in the efficiency of prompt tuning, determined by the norm of the \emph{soft-prompt-induced} keys and queries, and provide an upper bound criterion. Beyond this criterion, no sub-quadratic (efficient) algorithm for prompt tuning exists under SETH. Within this criterion, we showcase our theory by proving the existence of almost-linear time prompt tuning inference algorithms. These fundamental limits provide important necessary conditions for designing expressive and efficient prompt tuning methods for practitioners.
format Preprint
id arxiv_https___arxiv_org_abs_2411_16525
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Fundamental Limits of Prompt Tuning Transformers: Universality, Capacity and Efficiency
Hu, Jerry Yao-Chieh
Wang, Wei-Po
Gilani, Ammar
Li, Chenyang
Song, Zhao
Liu, Han
Machine Learning
Artificial Intelligence
Computation and Language
We investigate the statistical and computational limits of prompt tuning for transformer-based foundation models. Our key contributions are prompt tuning on \emph{single-head} transformers with only a \emph{single} self-attention layer: (i) is universal, and (ii) supports efficient (even almost-linear time) algorithms under the Strong Exponential Time Hypothesis (SETH). Statistically, we prove that prompt tuning on such simplest possible transformers are universal approximators for sequence-to-sequence Lipschitz functions. In addition, we provide an exponential-in-$dL$ and -in-$(1/ε)$ lower bound on the required soft-prompt tokens for prompt tuning to memorize any dataset with 1-layer, 1-head transformers. Computationally, we identify a phase transition in the efficiency of prompt tuning, determined by the norm of the \emph{soft-prompt-induced} keys and queries, and provide an upper bound criterion. Beyond this criterion, no sub-quadratic (efficient) algorithm for prompt tuning exists under SETH. Within this criterion, we showcase our theory by proving the existence of almost-linear time prompt tuning inference algorithms. These fundamental limits provide important necessary conditions for designing expressive and efficient prompt tuning methods for practitioners.
title Fundamental Limits of Prompt Tuning Transformers: Universality, Capacity and Efficiency
topic Machine Learning
Artificial Intelligence
Computation and Language
url https://arxiv.org/abs/2411.16525