Pretrained Transformers as Universal Computation Engines

by   Kevin Lu, et al.

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 improves performance and compute efficiency on non-language downstream tasks. In particular, we find that such pretraining enables FPT to generalize in zero-shot to these modalities, matching the performance of a transformer fully trained on these tasks.


page 1

page 7


Knowledge-Aware Language Model Pretraining

How much knowledge do pretrained language models hold? Recent research o...

What Happens During Finetuning of Vision Transformers: An Invariance Based Investigation

The pretrain-finetune paradigm usually improves downstream performance o...

NormFormer: Improved Transformer Pretraining with Extra Normalization

During pretraining, the Pre-LayerNorm transformer suffers from a gradien...

Pretraining task diversity and the emergence of non-Bayesian in-context learning for regression

Pretrained transformers exhibit the remarkable ability of in-context lea...

Learning Probabilistic Symmetrization for Architecture Agnostic Equivariance

We present a novel framework to overcome the limitations of equivariant ...

Finetuning Pretrained Transformers into RNNs

Transformers have outperformed recurrent neural networks (RNNs) in natur...

Code Repositories


Official codebase for Pretrained Transformers as Universal Computation Engines.

view repo

Please sign up or login with your details

Forgot password? Click here to reset