Characterizing Communication in Distributed Parameter-Efficient Fine-Tuning for LLMs

Watch a 29-minute technical presentation from the HOTI Hot Interconnects Symposium exploring the communication patterns and characteristics in distributed Parameter-Efficient Fine-Tuning (PEFT) approaches for Large Language Models. Led by researchers Nawras Alnaasan, Horng-Ruey Huang, Aamir Shafi, Hari Subramoni and Dhabaleswar K. Panda, and chaired by AMD's Shelby Lockhart, learn about the networking and interconnect challenges involved in efficiently fine-tuning massive language models across distributed systems. Gain insights into optimizing communication overhead and scaling PEFT methods as part of the Networks for Large Language Models technical paper session.