_version_ 1866909606165348352
author Lee, Yejin
Sun, Anna
Hosmer, Basil
Acun, Bilge
Balioglu, Can
Wang, Changhan
Hernandez, Charles David
Puhrsch, Christian
Haziza, Daniel
Guessous, Driss
Massa, Francisco
Kahn, Jacob
Wan, Jeffrey
Reizenstein, Jeremy
Zhai, Jiaqi
Isaacson, Joe
Schlosser, Joel
Pino, Juan
Sadagopan, Kaushik Ram
Shamis, Leonid
Ma, Linjian
Hwang, Min-Jae
Chen, Mingda
Elhoushi, Mostafa
Rodriguez, Pedro
Pasunuru, Ram
Yih, Scott
Popuri, Sravya
Liu, Xing
Wu, Carole-Jean
author_facet Lee, Yejin
Sun, Anna
Hosmer, Basil
Acun, Bilge
Balioglu, Can
Wang, Changhan
Hernandez, Charles David
Puhrsch, Christian
Haziza, Daniel
Guessous, Driss
Massa, Francisco
Kahn, Jacob
Wan, Jeffrey
Reizenstein, Jeremy
Zhai, Jiaqi
Isaacson, Joe
Schlosser, Joel
Pino, Juan
Sadagopan, Kaushik Ram
Shamis, Leonid
Ma, Linjian
Hwang, Min-Jae
Chen, Mingda
Elhoushi, Mostafa
Rodriguez, Pedro
Pasunuru, Ram
Yih, Scott
Popuri, Sravya
Liu, Xing
Wu, Carole-Jean
contents Generative artificial intelligence (AI) technology is revolutionizing the computing industry. Not only its applications have broadened to various sectors but also poses new system design and optimization opportunities. The technology is capable of understanding and responding in multiple modalities. However, the advanced capability currently comes with significant system resource demands. To sustainably scale generative AI capabilities to billions of users in the world, inference must be fast and efficient. This paper pinpoints key system design and optimization opportunities by characterizing a family of emerging multi-modal generation models on real systems. Auto-regressive token generation is a critical latency performance bottleneck, typically dominated by GPU idle time. In addition to memory-intensive attention across the generative AI models, linear operations constitute significant inference latency due to the feed forward networks in Transformer-based models. We demonstrate that state-of-the-art optimization levers, spanning from applications to system software and hardware, set a 3.88x better baseline.
format Preprint
id arxiv_https___arxiv_org_abs_2410_00215
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Characterizing and Efficiently Accelerating Multimodal Generation Model Inference
Lee, Yejin
Sun, Anna
Hosmer, Basil
Acun, Bilge
Balioglu, Can
Wang, Changhan
Hernandez, Charles David
Puhrsch, Christian
Haziza, Daniel
Guessous, Driss
Massa, Francisco
Kahn, Jacob
Wan, Jeffrey
Reizenstein, Jeremy
Zhai, Jiaqi
Isaacson, Joe
Schlosser, Joel
Pino, Juan
Sadagopan, Kaushik Ram
Shamis, Leonid
Ma, Linjian
Hwang, Min-Jae
Chen, Mingda
Elhoushi, Mostafa
Rodriguez, Pedro
Pasunuru, Ram
Yih, Scott
Popuri, Sravya
Liu, Xing
Wu, Carole-Jean
Machine Learning
Generative artificial intelligence (AI) technology is revolutionizing the computing industry. Not only its applications have broadened to various sectors but also poses new system design and optimization opportunities. The technology is capable of understanding and responding in multiple modalities. However, the advanced capability currently comes with significant system resource demands. To sustainably scale generative AI capabilities to billions of users in the world, inference must be fast and efficient. This paper pinpoints key system design and optimization opportunities by characterizing a family of emerging multi-modal generation models on real systems. Auto-regressive token generation is a critical latency performance bottleneck, typically dominated by GPU idle time. In addition to memory-intensive attention across the generative AI models, linear operations constitute significant inference latency due to the feed forward networks in Transformer-based models. We demonstrate that state-of-the-art optimization levers, spanning from applications to system software and hardware, set a 3.88x better baseline.
title Characterizing and Efficiently Accelerating Multimodal Generation Model Inference
topic Machine Learning
url https://arxiv.org/abs/2410.00215