AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

Transformers are RNNs: Fast Autoregressive Transformers with Linear Attention

Angelos Katharopoulos, Apoorv Vyas, Nikolaos Pappas, François Fleuret

ICML 2020 · Architectures and efficiency › Attention and transformers · approach: Sparse efficient transformers for long context · 301 citations (OpenAlex)

Read the paper

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
Replace softmax attention with positive feature-map linear attentionAutoregressive image generation · MNISTimages/sec: 0.45 -> 142.8Improved
Replace softmax attention with positive feature-map linear attentionAutoregressive image generation · MNISTbits/dim: 0.621 -> 0.644No clear effect
Replace softmax attention with positive feature-map linear attentionAutoregressive image generation · CIFAR-10images/sec: 0.004 -> 17.85Improved
Replace softmax attention with positive feature-map linear attentionAutoregressive image generation · CIFAR-10bits/dim: 3.47 -> 3.40Improved
Replace softmax attention with positive feature-map linear attentionTransformer · WSJValidation PER: 5.12 -> 8.08Hurt
Replace softmax attention with positive feature-map linear attentionTransformer · WSJTime/epoch (s): 2711 -> 824Improved
Use linear attention instead of LSH attentionAutoregressive image generation · MNISTimages/sec: 0.68 -> 142.8Improved
Use linear attention instead of LSH attentionTransformer · WSJValidation PER: 9.33 -> 8.08Improved
Cache and reuse softmax keys and values across autoregressive stepsSoftmax transformer · MNIST image generationimages/sec: 0.45 -> 7.56Improved
Use linear attention instead of cached softmax attentionAutoregressive image generation · CIFAR-10images/sec: 0.32 -> 17.85Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy