MA6b.3

An In-Memory Design of Byte-Pair Encoding Tokenization for Large Language Models

Eros Allan Hacinas, Charibeth Cheng, De La Salle University, Philippines; Gregory Kielian, Google, United States; Ann Franchesca Laguna, De La Salle University, Philippines
Session:
MA6b: Emerging Technologies for Artificial Intelligence Systems Lecture
Track:
Architectures and Implementation
Location:
Room 6
Presentation Time:
Mon, 26 Oct, 11:05 - 11:30 PT (UTC -7)
Session Co-Chairs:
Dayane Reis, University of South Florida, United States and Zheyu Yan, Zhejiang University, China
Presentation
Discussion
Resources
No resources available.
Session MA6b
MA6b.1: On the Viability of Content-Addressable Memory for Retrieval-Augmented Generation
Yuanjie Zhang, Yongtao Zhou, Zheyu Yan, Zhejiang University, China
MA6b.2: Hardware–Software Co-Design of Near-Zero Multiplication Escape for Sparse Neural Networks
Parsa Khorrami, John Maurer, Dayane Reis, University of South Florida, United States
MA6b.3: An In-Memory Design of Byte-Pair Encoding Tokenization for Large Language Models
Eros Allan Hacinas, Charibeth Cheng, De La Salle University, Philippines; Gregory Kielian, Google, United States; Ann Franchesca Laguna, De La Salle University, Philippines
MA6b.4: Enable Acoustic Imaging System Designs with Quantum Optimization
Hanhan Wu, Yipei Liu, George Mason University, United States; Zebo Yang, Florida Atlantic University, United States; Youzuo Lin, University of North Carolina at Chapel Hill, United States; Weiwen Jiang, Lei Yang, George Mason University, United States
Contacts