Pretrained Transformers as Universal Computation Engines

TL;DR

This paper explores how pretrained language transformers can be adapted to perform well on diverse non-language tasks with minimal modifications, demonstrating their versatility as universal computation engines.

Contribution

It introduces the concept of Frozen Pretrained Transformers (FPT) and shows their effectiveness across multiple modalities without changing self-attention or feedforward layers.

Findings

Pretraining on language improves non-language task performance.

FPT achieves strong results with minimal finetuning.

Language-pretrained transformers outperform random initialized models.

Abstract

We investigate the capability of a transformer pretrained on natural language to generalize to other modalities with minimal finetuning -- in particular, without finetuning of the self-attention and feedforward layers of the residual blocks. We consider such a model, which we call a Frozen Pretrained Transformer (FPT), and study finetuning it on a variety of sequence classification tasks spanning numerical computation, vision, and protein fold prediction. In contrast to prior works which investigate finetuning on the same modality as the pretraining dataset, we show that pretraining on natural language can improve performance and compute efficiency on non-language downstream tasks. Additionally, we perform an analysis of the architecture, comparing the performance of a random initialized transformer to a random LSTM. Combining the two insights, we find language-pretrained transformers can obtain strong performance on a variety of non-language tasks.