Long-Short Transformer: Efficient Transformers for Language and Vision

Abstract

Transformers have achieved success in both language and vision domains. However, it is prohibitively expensive to scale them to long sequences such as long documents or high-resolution images, because self-attention mechanism has quadratic time and memory complexities with respect to the input sequence length. In this paper, we propose Long-Short Transformer (Transformer-LS), an efficient self-attention mechanism for modeling long sequences with linear complexity for both language and vision tasks. It aggregates a novel long-range attention with dynamic projection to model distant correlations and a short-term attention to capture fine-grained local correlations. We propose a dual normalization strategy to account for the scale mismatch between the two attention mechanisms. Transformer-LS can be applied to both autoregressive and bidirectional models without additional complexity. Our method outperforms the state-of-the-art models on multiple tasks in language and vision domains, including the Long Range Arena benchmark, autoregressive language modeling, and ImageNet classification. For instance, Transformer-LS achieves 0.97 test BPC on enwik8 using half the number of parameters than previous method, while being faster and is able to handle 3x as long sequences compared to its full-attention version on the same hardware. On ImageNet, it can obtain the state-of-the-art results (e.g., a moderate size of 55.8M model solely trained on 224x224 ImageNet-1K can obtain Top-1 accuracy 84.1%), while being more scalable on high-resolution images. The source code and models are released at https://github.com/NVIDIA/transformer-ls .

Published at NeurIPS 2021

References in corpus (19)

• Language Models are Few-Shot Learners
• PVT v2: Improved Baselines with Pyramid Vision Transformer
• Zero-Shot Text-to-Image Generation
• Linformer: Self-Attention with Linear Complexity
• Generating Long Sequences with Sparse Transformers
• Do ImageNet Classifiers Generalize to ImageNet?
• Axial Attention in Multidimensional Transformers
• Reformer: The Efficient Transformer
• Transformers are RNNs: Fast Autoregressive Transformers with Linear Attention
• CvT: Introducing Convolutions to Vision Transformers
• Lite Transformer with Long-Short Range Attention
• Perceiver: General Perception with Iterative Attention
• Random Feature Attention
• Generating Diverse High-Fidelity Images with VQ-VAE-2
• Sparse Sinkhorn Attention
• Are we done with ImageNet?
• Noise or Signal: The Role of Image Backgrounds in Object Recognition
• BP-Transformer: Modelling Long-Range Context via Binary Partitioning
• Vision Transformers with Patch Diversification

Cited by in corpus (7)

• LinRec: Linear Attention Mechanism for Long-term Sequential Recommender Systems
• AS-MLP: An Axial Shifted MLP Architecture for Vision
• Long-Range Transformers for Dynamic Spatiotemporal Forecasting
• Adaptive Fourier Neural Operators: Efficient Token Mixers for Transformers
• Fourier Transformer: Fast Long Range Modeling by Removing Sequence Redundancy with FFT Operator
• Scatterbrain: Unifying Sparse and Low-rank Attention Approximation
• Sparse Factorization of Large Square Matrices