Skip to main content
PaperLLMsv1.0

Transformers are RNNs: Fast Autoregressive Transformers with Linear Attention

by Idiap Research Institute / EPFL · free · Last verified 2026-03-17

Shows that by approximating the softmax attention kernel, transformers can be expressed as linear RNNs, enabling O(1) autoregressive inference. Introduces the linear attention framework that inspired many subsequent efficient attention variants.

https://arxiv.org/abs/2006.16236
C+
C+Average
Adoption: BQuality: AFreshness: CCitations: BEngagement: F

Specifications

License
Open Access
Pricing
free
Capabilities
linear-attention, efficient-inference, rnn-equivalence
Integrations
Use Cases
efficient-sequence-modeling, real-time-inference
API Available
No
Tags
linear-attention, rnn-equivalence, efficient-transformers, kernel-approximation, inference
Added
2026-03-17
Completeness
100%

Index Score

58
Adoption
62
Quality
86
Freshness
42
Citations
64
Engagement
0

Put AI to work for your business

Deploy this paper alongside autonomous AaaS agents that handle tasks end-to-end — no babysitting required.

Explore the full AI ecosystem on Agents as a Service