Skip to content

Attention is all you need

Authors: Ashish Vaswani; Noam Shazeer; Niki Parmar; Jakob Uszkoreit; Llion Jones; Aidan N. Gomez; Lukasz Kaiser; Illia Polosukhin
Year: 2017
Venue: arXiv preprint

type: other collection: sok-references status: auto

Links: arXiv:1706.03762

Summary

Auto-extracted summary (unreviewed)

Introduces the Transformer, a sequence-transduction architecture built entirely from multi-head self-attention and feed-forward layers, with no recurrence or convolution, which is more parallelizable and cheaper to train than RNN/CNN encoder-decoders. The big model reaches 28.4 BLEU on WMT 2014 English-German (over 2 BLEU above prior ensembles) and a new single-model state of the art on WMT 2014 English-French (41.8 BLEU in the abstract, 41.0 in the results section) after 3.5 days on eight P100 GPUs, and also transfers to English constituency parsing.

Abstract

The dominant sequence transduction models are based on complex recurrent or convolutional neural networks in an encoder-decoder configuration. The best performing models also connect the encoder and decoder through an attention mechanism. We propose a new simple network architecture, the Transformer, based solely on attention mechanisms, dispensing with recurrence and convolutions entirely. Experiments on two machine translation tasks show these models to be superior in quality while being more parallelizable and requiring significantly less time to train. Our model achieves 28.4 BLEU on the WMT 2014 English-to-German translation task, improving over the existing best results, including ensembles by over 2 BLEU. On the WMT 2014 English-to-French translation task, our model establishes a new single-model state-of-the-art BLEU score of 41.8 after training for 3.5 days on eight GPUs, a small fraction of the training costs of the best models from the literature. We show that the Transformer generalizes well to other tasks by applying it successfully to English constituency parsing both with large and limited training data.
Source: arxiv

Extracted details

Field Value Source Confidence
contribution Transformer architecture based solely on attention llm 0.95
contribution scaled dot-product attention and multi-head attention llm 0.95
contribution sinusoidal positional encoding llm 0.95
contribution state-of-the-art translation quality at a fraction of prior training cost llm 0.95
contribution generalization to English constituency parsing llm 0.95
libraries tensor2tensor llm 0.95
models Transformer (base) llm 0.95
models Transformer (big) llm 0.95
datasets WMT 2014 English-German llm 0.95
datasets WMT 2014 English-French llm 0.95
datasets Penn Treebank WSJ llm 0.95
tasks machine translation llm 0.95
tasks constituency parsing llm 0.95
hardware 8x NVIDIA P100 GPU llm 0.95
code url https://github.com/tensorflow/tensor2tensor (URL found in the full text) regex 0.50
code url https://github.com/tensorflow/tensor2tensor llm 0.95
key result BLEU: 28.4 (WMT 2014 English-to-German, Transformer (big)) llm 0.95
key result BLEU: 41.8 (abstract) / 41.0 (results section) (WMT 2014 English-to-French, Transformer (big)) llm 0.95
key result training time: 3.5 days on 8 P100 GPUs (Transformer (big)) llm 0.95
key result parameters: 65M (base), 213M (big) (Table 3) llm 0.95

SoK reference [37]

A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, L. u. Kaiser, and I. Polosukhin, “Attention is all you need,” in Advances in Neural Information Processing Systems, I. Guyon, U. V. Luxburg, S. Bengio, H. Wallach, R. Fergus, S. Vishwanathan, and R. Garnett, Eds., vol. 30. Curran Associates, Inc., 2017. [Online]. Available: https://proceedings.neurips.cc/paper files/paper/ 2017/file/3f5ee243547dee91fbd053c1c4a845aa-Paper.pdf

Listed on the reference page of the site.

Keyword mentions

Counts of vocabulary terms in the full text (first page of occurrence in brackets).

  • model: Transformer 26 [p.1], RNN / LSTM 11 [p.1], CNN 2 [p.1], Mixture of experts 2 [p.8]
  • dataset: Penn Treebank 1 [p.9]
  • hardware: GPU 8 [p.1]
  • technique: Attention 42 [p.1], Convolution 16 [p.1], Softmax 7 [p.4], LayerNorm 4 [p.3]
  • task: Machine translation 12 [p.1], Language modeling / generation 4 [p.2]

Files

Path (under papers/) Collection Pages Size SHA-256 Note
references/sok-references/037_attention-is-all-you-need.pdf sok-references 15 2.1 MB bdfaa68d8984

Database key: attention-is-all-you-need · created 2026-11-05 · updated 2026-11-05 · edit with python3 tools/paperdb.py set attention-is-all-you-need COLUMN VALUE