Provided by

arXiv

arxiv.org
𝕏gh
0.0/5
0 views
0 saved
FreeadvancedText Generation
ResearchersAI EngineersDevelopersStudents
Introduces the Transformer architecture, a novel neural network model based solely on self-attention mechanisms, revolutionizing sequence transduct...

Overview

This seminal paper introduces the Transformer, a groundbreaking neural network architecture that eschews recurrence and convolutions entirely, relying instead on a self-attention mechanism. It demonstrates superior performance on machine translation tasks, significantly reducing training time and paving the way for large language models. The Transformer's parallelizable design and ability to capture long-range dependencies efficiently have made it the dominant architecture in Natural Language Processing (NLP) and increasingly in other domains like computer vision, profoundly shaping the landscape of AI research and applications.

Abstract

The dominant sequence transduction models are based on complex recurrent or convolutional neural networks that include an encoder-decoder. The best performing models also connect the encoder and decoder with an attention mechanism. We propose a new simple network architecture, the Transformer, based solely on attention mechanisms, eschewing 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 best previously reported ensembles by over 2 BLEU, including on the training time. 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 cost of the best models from the literature. We show that the Transformer generalizes well to other tasks by applying it to English constituency parsing, both with large and limited training data.