June 03, 2024
Large Language Models (LLMs) with hundreds of billions of parameters have transformed the field of machine learning. However, serving these models at inference time is both compute and memory intensive, where a single request can require multiple GPUs and tens of Gigabytes of memory. Multi-Head Attention is one of the key components of LLMs, which can account for over 50% of LLMs memory and compute requirement. We observe that there is a high amount of redundancy across heads on which tokens they pay attention to. Based on this insight, we propose Clustered Head Attention ( CHAI ). CHAI combines heads with a high amount of correlation for self-attention at runtime, thus reducing both memory and compute. In our experiments, we show that CHAI is able to reduce the memory requirements for storing K,V cache by up to 21.4% and inference time latency by up to 1.73× without any fine-tuning required. CHAI achieves this with a maximum 3.2% deviation in accuracy across 3 different models (i.e. OPT-66B, LLAMA-7B, LLAMA-33B) and 5 different evaluation datasets.
Publisher
ICML
Research Topics
November 20, 2024
Jay Shah, Ganesh Bikshandi, Ying Zhang, Vijay Thakkar, Pradeep Ramani, Tri Dao
November 20, 2024
July 23, 2024
Shengye Wan, Cyrus Nikolaidis, Daniel Song, David Molnar, James Crnkovich, Jayson Grace, Manish Bhatt, Sahana Chennabasappa, Spencer Whitman, Stephanie Ding, Vlad Ionescu, Yue Li, Joshua Saxe
July 23, 2024
June 27, 2024
Chris Cummins, Volker Seeker, Dejan Grubisic, Baptiste Rozière, Jonas Gehring, Gabriel Synnaeve, Hugh Leather
June 27, 2024
June 14, 2024
Mostafa Elhoushi, Akshat Shrivastava, Diana Liskovich, Basil Hosmer, Bram Wasti, Liangzhen Lai, Bilge Acun, Ahmed Aly, Beidi Chen, Carole-Jean Wu, Ahmed Roman, Nas Mahmoud, Saurabh Agarwal
June 14, 2024
Foundational models
Latest news
Foundational models