Vaswani et al. • 2017

Abstract

This landmark paper introduces the Transformer architecture, derived entirely from attention mechanisms, dispensing with recurrence and convolutions. It solved the problem of parallelization in sequence processing, becoming the foundational architecture for virtually all modern Large Language Models.

Why It Matters

• Introduced the self-attention mechanism
• Enabled massive parallel training
• Foundation for GPT, Claude, Llama, and other major models