Building a transformer with PyTorch

• Titel: Building a transformer with PyTorch
• Mitwirkende: Hermans, Jeroen [MitwirkendeR]
• Institutionen: Packt Publishing, [Verlag]
• Ausgabe: [First edition].
• Verlagsort: [Birmingham, United Kingdom]
• Verlag: Packt Publishing
• E-Jahr: 2025
• Jahr: [2025]
• Umfang: 1 online resource (1 video file (2 hr., 56 min.))
• Illustrationen: sound, color.
• Fussnoten: "DataLab" : DataLab is a learning solution designed to empower individuals and teams to explore data, build skills, and solve real-world problems. With interactive content, hands-on exercises, and a focus on practical application, DataLab helps learners turn data curiosity into confident capability—no matter where they start. - Online resource; title from title details screen (O’Reilly, viewed April 9, 2025)
• ISBN: 978-1-80602-155-0
• ISBN: 1-80602-155-2
• Datenträger: Online-Ressource
• Sprache: eng
• Sach-SW: Traitement automatique des langues naturelles
• Sach-SW: Réseaux neuronaux (Informatique)
• Sach-SW: Python (Langage de programmation)
• Sach-SW: Intelligence artificielle
• Sach-SW: artificial intelligence
• Form-SW: Instructional films
• Form-SW: Nonfiction films
• Form-SW: Internet videos
• Form-SW: Films de formation
• Form-SW: Films autres que de fiction
• Form-SW: Vidéos sur Internet
• K10plus-PPN: 1924460106
• K10plus-PPN:
  Universitätsbibliothek
• Lokale URL UB: Zum Volltext
• Lokale URL UB: Bibliothek der Medizinischen Fakultät Mannheim der Universität Heidelberg
• Bibliothek/Idn: UW / m471896024X
• Lokale URL Inst.: Zum Volltext

Abstract

Transformers have revolutionized the field of AI, especially in natural language processing, powering models like GPT and BERT. In this code-along, you'll dive deep into the architecture that started it all—the Transformer, introduced in the seminal paper Attention Is All You Need. Through a hands-on PyTorch implementation, you'll gain a clear understanding of key components like self-attention, positional encoding, multi-head attention, and feedforward layers. Each part of the architecture is broken down and explained, allowing you to build a working transformer model from the ground up. By applying the model to a simple use case, you'll see how transformers process sequences in parallel, enabling faster and more efficient learning than traditional RNNs. Whether you're aiming to understand how large language models work or want to build your own, this session offers a solid foundation in modern deep learning.