Explore co-design methodologies for distributed machine learning platforms through a comprehensive conference talk from Hot Interconnects Symposium featuring Georgia Tech researchers Tushar Krishna and William Won. Dive into the technical aspects of ASTRA-sim and Chakra frameworks, examining their roles in optimizing distributed ML systems. Under the session chair guidance of Intel's David Ozog, learn about innovative approaches to platform design and performance optimization for large-scale machine learning implementations.
ASTRA-sim and Chakra - Co-design Exploration for Distributed Machine Learning Platforms
HOTI - Hot Interconnects Symposium via YouTube
Overview
Syllabus
Day 3 15:00: ASTRA-sim and Chakra: Co-design Exploration for Distributed Machine Learning Platforms
Taught by
HOTI - Hot Interconnects Symposium
Related Courses
-
Distributed Optimization and Machine Learning
-
Towards a Standardized Representation for Deep Learning Collective Algorithms
-
Principles and Practice of Scalable and Distributed Deep Neural Networks Training and Inference
-
Demystifying the Communication Characteristics for Distributed Transformer Models
-
Keeping Pace: Micro-Models, Large Language Models, and the Co-Evolution of Chip and Data Center Design in a Rapidly Shifting AI Landscape
-
High Performance Machine Learning, Deep Learning, and Data Science - Principles and Practice
