Copertina

External Large Foundation Model: How to Efficiently Serve Trillions of Parameters for Online Ads Recommendation

Fuente: arXiv

Salvato in:

Dettagli Bibliografici
Autori principali:	Liang, Mingfu, Liu, Xi, Jin, Rong, Liu, Boyang, Suo, Qiuling, Zhou, Qinghai, Zhou, Song, Chen, Laming, Zheng, Hua, Li, Zhiyuan, Jiang, Shali, Yang, Jiyan, Xia, Xiaozhen, Yang, Fan, Badr, Yasmine, Wen, Ellie, Xu, Shuyu, Chen, Hansey, Zhang, Zhengyu, Nie, Jade, Yang, Chunzhi, Zeng, Zhichen, Zhang, Weilin, Huang, Xingliang, Li, Qianru, Wang, Shiquan, Lyu, Evelyn, Lu, Wenjing, Zhang, Rui, Wang, Wenjun, Rudy, Jason, Hang, Mengyue, Wang, Kai, Ma, Yinbin, Wang, Shuaiwen, Zeng, Sihan, Tang, Tongyi, Wei, Xiaohan, Jin, Longhao, Zhang, Jamey, Chen, Marcus, Xu, Jiayi, Huang, Angie, Zeng, Xihuan, Zhang, Chi, Zhao, Zhengli, Yang, Jared, Jin, Qiang, Chen, Xian, Amlesahwaram, Amit Anand, Song, Lexi, Luo, Liang, Hao, Yuchen, Xiao, Nan, Yetim, Yavuz, Pan, Luoshang, Liu, Gaoxiang, Hu, Yuxi, Huang, Yuzhen, Xu, Jackie, Zhu, Rich, Zhang, Xin, Liu, Yiqun, Yin, Hang, Chen, Yuxin, Zhang, Buyun, Liu, Xiaoyi, Wang, Xingyuan, Mao, Wenguang, Li, Zhijing, Zhou, Zhehui, Gu, Feifan, Huang, Qin, Sun, Chonglin, Yu, Nancy, Gu, Shuo, Mao, Shupin, Au, Benjamin, Qin, Jingzheng, Yao, Peggy, Choi, Jae-Woo, Gao, Bin, Wang, Ernest, Zhang, Lei, Chen, Wen-Yen, Lee, Ted, Zha, Yujie, Meng, Yi, Gong, Alex, Gao, Edison, Hsueh, Jack, Zheng, Jie, Vahdatpour, Alireza, Han, Yiping, Yao, Yantao, Kureha, Toshinari, Chang, Shuo, Sultan, Musharaf, Bocharov, John, Chordia, Sagar, Gan, Xiaorui, Sun, Peng, Liu, Rocky, Long, Bo, Chen, Wenlin, Kolay, Santanu, Li, Huayu
Natura:	Preprint
Pubblicazione:	2025
Soggetti:	Information Retrieval Artificial Intelligence Machine Learning
Accesso online:	Acceder al recurso
Tags:	Aggiungi Tag Nessun Tag, puoi essere il primo ad aggiungerne!!

_version_	1866916841348136960
author	Liang, Mingfu Liu, Xi Jin, Rong Liu, Boyang Suo, Qiuling Zhou, Qinghai Zhou, Song Chen, Laming Zheng, Hua Li, Zhiyuan Jiang, Shali Yang, Jiyan Xia, Xiaozhen Yang, Fan Badr, Yasmine Wen, Ellie Xu, Shuyu Chen, Hansey Zhang, Zhengyu Nie, Jade Yang, Chunzhi Zeng, Zhichen Zhang, Weilin Huang, Xingliang Li, Qianru Wang, Shiquan Lyu, Evelyn Lu, Wenjing Zhang, Rui Wang, Wenjun Rudy, Jason Hang, Mengyue Wang, Kai Ma, Yinbin Wang, Shuaiwen Zeng, Sihan Tang, Tongyi Wei, Xiaohan Jin, Longhao Zhang, Jamey Chen, Marcus Xu, Jiayi Huang, Angie Zeng, Xihuan Zhang, Chi Zhao, Zhengli Yang, Jared Jin, Qiang Chen, Xian Amlesahwaram, Amit Anand Song, Lexi Luo, Liang Hao, Yuchen Xiao, Nan Yetim, Yavuz Pan, Luoshang Liu, Gaoxiang Hu, Yuxi Huang, Yuzhen Xu, Jackie Zhu, Rich Zhang, Xin Liu, Yiqun Yin, Hang Chen, Yuxin Zhang, Buyun Liu, Xiaoyi Wang, Xingyuan Mao, Wenguang Li, Zhijing Zhou, Zhehui Gu, Feifan Huang, Qin Sun, Chonglin Yu, Nancy Gu, Shuo Mao, Shupin Au, Benjamin Qin, Jingzheng Yao, Peggy Choi, Jae-Woo Gao, Bin Wang, Ernest Zhang, Lei Chen, Wen-Yen Lee, Ted Zha, Yujie Meng, Yi Gong, Alex Gao, Edison Hsueh, Jack Zheng, Jie Vahdatpour, Alireza Han, Yiping Yao, Yantao Kureha, Toshinari Chang, Shuo Sultan, Musharaf Bocharov, John Chordia, Sagar Gan, Xiaorui Sun, Peng Liu, Rocky Long, Bo Chen, Wenlin Kolay, Santanu Li, Huayu
author_facet	Liang, Mingfu Liu, Xi Jin, Rong Liu, Boyang Suo, Qiuling Zhou, Qinghai Zhou, Song Chen, Laming Zheng, Hua Li, Zhiyuan Jiang, Shali Yang, Jiyan Xia, Xiaozhen Yang, Fan Badr, Yasmine Wen, Ellie Xu, Shuyu Chen, Hansey Zhang, Zhengyu Nie, Jade Yang, Chunzhi Zeng, Zhichen Zhang, Weilin Huang, Xingliang Li, Qianru Wang, Shiquan Lyu, Evelyn Lu, Wenjing Zhang, Rui Wang, Wenjun Rudy, Jason Hang, Mengyue Wang, Kai Ma, Yinbin Wang, Shuaiwen Zeng, Sihan Tang, Tongyi Wei, Xiaohan Jin, Longhao Zhang, Jamey Chen, Marcus Xu, Jiayi Huang, Angie Zeng, Xihuan Zhang, Chi Zhao, Zhengli Yang, Jared Jin, Qiang Chen, Xian Amlesahwaram, Amit Anand Song, Lexi Luo, Liang Hao, Yuchen Xiao, Nan Yetim, Yavuz Pan, Luoshang Liu, Gaoxiang Hu, Yuxi Huang, Yuzhen Xu, Jackie Zhu, Rich Zhang, Xin Liu, Yiqun Yin, Hang Chen, Yuxin Zhang, Buyun Liu, Xiaoyi Wang, Xingyuan Mao, Wenguang Li, Zhijing Zhou, Zhehui Gu, Feifan Huang, Qin Sun, Chonglin Yu, Nancy Gu, Shuo Mao, Shupin Au, Benjamin Qin, Jingzheng Yao, Peggy Choi, Jae-Woo Gao, Bin Wang, Ernest Zhang, Lei Chen, Wen-Yen Lee, Ted Zha, Yujie Meng, Yi Gong, Alex Gao, Edison Hsueh, Jack Zheng, Jie Vahdatpour, Alireza Han, Yiping Yao, Yantao Kureha, Toshinari Chang, Shuo Sultan, Musharaf Bocharov, John Chordia, Sagar Gan, Xiaorui Sun, Peng Liu, Rocky Long, Bo Chen, Wenlin Kolay, Santanu Li, Huayu
contents	Ads recommendation is a prominent service of online advertising systems and has been actively studied. Recent studies indicate that scaling-up and advanced design of the recommendation model can bring significant performance improvement. However, with a larger model scale, such prior studies have a significantly increasing gap from industry as they often neglect two fundamental challenges in industrial-scale applications. First, training and inference budgets are restricted for the model to be served, exceeding which may incur latency and impair user experience. Second, large-volume data arrive in a streaming mode with data distributions dynamically shifting, as new users/ads join and existing users/ads leave the system. We propose the External Large Foundation Model (ExFM) framework to address the overlooked challenges. Specifically, we develop external distillation and a data augmentation system (DAS) to control the computational cost of training/inference while maintaining high performance. We design the teacher in a way like a foundation model (FM) that can serve multiple students as vertical models (VMs) to amortize its building cost. We propose Auxiliary Head and Student Adapter to mitigate the data distribution gap between FM and VMs caused by the streaming data issue. Comprehensive experiments on internal industrial-scale applications and public datasets demonstrate significant performance gain by ExFM.
format	Preprint
id	arxiv_https___arxiv_org_abs_2502_17494
institution	arXiv
publishDate	2025
record_format	arxiv
spellingShingle	External Large Foundation Model: How to Efficiently Serve Trillions of Parameters for Online Ads Recommendation Liang, Mingfu Liu, Xi Jin, Rong Liu, Boyang Suo, Qiuling Zhou, Qinghai Zhou, Song Chen, Laming Zheng, Hua Li, Zhiyuan Jiang, Shali Yang, Jiyan Xia, Xiaozhen Yang, Fan Badr, Yasmine Wen, Ellie Xu, Shuyu Chen, Hansey Zhang, Zhengyu Nie, Jade Yang, Chunzhi Zeng, Zhichen Zhang, Weilin Huang, Xingliang Li, Qianru Wang, Shiquan Lyu, Evelyn Lu, Wenjing Zhang, Rui Wang, Wenjun Rudy, Jason Hang, Mengyue Wang, Kai Ma, Yinbin Wang, Shuaiwen Zeng, Sihan Tang, Tongyi Wei, Xiaohan Jin, Longhao Zhang, Jamey Chen, Marcus Xu, Jiayi Huang, Angie Zeng, Xihuan Zhang, Chi Zhao, Zhengli Yang, Jared Jin, Qiang Chen, Xian Amlesahwaram, Amit Anand Song, Lexi Luo, Liang Hao, Yuchen Xiao, Nan Yetim, Yavuz Pan, Luoshang Liu, Gaoxiang Hu, Yuxi Huang, Yuzhen Xu, Jackie Zhu, Rich Zhang, Xin Liu, Yiqun Yin, Hang Chen, Yuxin Zhang, Buyun Liu, Xiaoyi Wang, Xingyuan Mao, Wenguang Li, Zhijing Zhou, Zhehui Gu, Feifan Huang, Qin Sun, Chonglin Yu, Nancy Gu, Shuo Mao, Shupin Au, Benjamin Qin, Jingzheng Yao, Peggy Choi, Jae-Woo Gao, Bin Wang, Ernest Zhang, Lei Chen, Wen-Yen Lee, Ted Zha, Yujie Meng, Yi Gong, Alex Gao, Edison Hsueh, Jack Zheng, Jie Vahdatpour, Alireza Han, Yiping Yao, Yantao Kureha, Toshinari Chang, Shuo Sultan, Musharaf Bocharov, John Chordia, Sagar Gan, Xiaorui Sun, Peng Liu, Rocky Long, Bo Chen, Wenlin Kolay, Santanu Li, Huayu Information Retrieval Artificial Intelligence Machine Learning Ads recommendation is a prominent service of online advertising systems and has been actively studied. Recent studies indicate that scaling-up and advanced design of the recommendation model can bring significant performance improvement. However, with a larger model scale, such prior studies have a significantly increasing gap from industry as they often neglect two fundamental challenges in industrial-scale applications. First, training and inference budgets are restricted for the model to be served, exceeding which may incur latency and impair user experience. Second, large-volume data arrive in a streaming mode with data distributions dynamically shifting, as new users/ads join and existing users/ads leave the system. We propose the External Large Foundation Model (ExFM) framework to address the overlooked challenges. Specifically, we develop external distillation and a data augmentation system (DAS) to control the computational cost of training/inference while maintaining high performance. We design the teacher in a way like a foundation model (FM) that can serve multiple students as vertical models (VMs) to amortize its building cost. We propose Auxiliary Head and Student Adapter to mitigate the data distribution gap between FM and VMs caused by the streaming data issue. Comprehensive experiments on internal industrial-scale applications and public datasets demonstrate significant performance gain by ExFM.
title	External Large Foundation Model: How to Efficiently Serve Trillions of Parameters for Online Ads Recommendation
topic	Information Retrieval Artificial Intelligence Machine Learning
url	https://arxiv.org/abs/2502.17494