{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,2,1]],"date-time":"2026-02-01T05:45:59Z","timestamp":1769924759337,"version":"3.49.0"},"reference-count":65,"publisher":"Institute of Electrical and Electronics Engineers (IEEE)","issue":"9","license":[{"start":{"date-parts":[[2025,9,1]],"date-time":"2025-09-01T00:00:00Z","timestamp":1756684800000},"content-version":"vor","delay-in-days":0,"URL":"https:\/\/ieeexplore.ieee.org\/Xplorehelp\/downloads\/license-information\/IEEE.html"},{"start":{"date-parts":[[2025,9,1]],"date-time":"2025-09-01T00:00:00Z","timestamp":1756684800000},"content-version":"stm-asf","delay-in-days":0,"URL":"https:\/\/doi.org\/10.15223\/policy-029"},{"start":{"date-parts":[[2025,9,1]],"date-time":"2025-09-01T00:00:00Z","timestamp":1756684800000},"content-version":"stm-asf","delay-in-days":0,"URL":"https:\/\/doi.org\/10.15223\/policy-037"}],"funder":[{"DOI":"10.13039\/501100001809","name":"National Natural Science Foundation of China","doi-asserted-by":"publisher","award":["62025208"],"award-info":[{"award-number":["62025208"]}],"id":[{"id":"10.13039\/501100001809","id-type":"DOI","asserted-by":"publisher"}]},{"DOI":"10.13039\/501100001809","name":"National Natural Science Foundation of China","doi-asserted-by":"publisher","award":["62421002"],"award-info":[{"award-number":["62421002"]}],"id":[{"id":"10.13039\/501100001809","id-type":"DOI","asserted-by":"publisher"}]}],"content-domain":{"domain":[],"crossmark-restriction":false},"short-container-title":["IEEE Trans. Parallel Distrib. Syst."],"published-print":{"date-parts":[[2025,9]]},"DOI":"10.1109\/tpds.2025.3583165","type":"journal-article","created":{"date-parts":[[2025,6,25]],"date-time":"2025-06-25T14:25:12Z","timestamp":1750861512000},"page":"1828-1840","source":"Crossref","is-referenced-by-count":1,"title":["Oases: Efficient Large-Scale Model Training on Commodity Servers via Overlapped and Automated Tensor Model Parallelism"],"prefix":"10.1109","volume":"36","author":[{"ORCID":"https:\/\/orcid.org\/0000-0002-7419-1511","authenticated-orcid":false,"given":"Shengwei","family":"Li","sequence":"first","affiliation":[{"name":"National Key Laboratory of Parallel and Distributed Computing, College of Computer Science and Technology, National University of Defense Technology, Changsha, Hunan, China"}],"role":[{"role":"author","vocabulary":"crossref"}]},{"ORCID":"https:\/\/orcid.org\/0000-0002-3458-4732","authenticated-orcid":false,"given":"Zhiquan","family":"Lai","sequence":"additional","affiliation":[{"name":"National Key Laboratory of Parallel and Distributed Computing, College of Computer Science and Technology, National University of Defense Technology, Changsha, Hunan, China"}],"role":[{"role":"author","vocabulary":"crossref"}]},{"ORCID":"https:\/\/orcid.org\/0000-0001-9743-2034","authenticated-orcid":false,"given":"Dongsheng","family":"Li","sequence":"additional","affiliation":[{"name":"National Key Laboratory of Parallel and Distributed Computing, College of Computer Science and Technology, National University of Defense Technology, Changsha, Hunan, China"}],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"Yanqi","family":"Hao","sequence":"additional","affiliation":[{"name":"National Key Laboratory of Parallel and Distributed Computing, College of Computer Science and Technology, National University of Defense Technology, Changsha, Hunan, China"}],"role":[{"role":"author","vocabulary":"crossref"}]},{"ORCID":"https:\/\/orcid.org\/0000-0002-0354-9467","authenticated-orcid":false,"given":"Weijie","family":"Liu","sequence":"additional","affiliation":[{"name":"National Key Laboratory of Parallel and Distributed Computing, College of Computer Science and Technology, National University of Defense Technology, Changsha, Hunan, China"}],"role":[{"role":"author","vocabulary":"crossref"}]},{"ORCID":"https:\/\/orcid.org\/0000-0002-0669-6892","authenticated-orcid":false,"given":"Keshi","family":"Ge","sequence":"additional","affiliation":[{"name":"National Key Laboratory of Parallel and Distributed Computing, College of Computer Science and Technology, National University of Defense Technology, Changsha, Hunan, China"}],"role":[{"role":"author","vocabulary":"crossref"}]},{"ORCID":"https:\/\/orcid.org\/0000-0003-0622-1202","authenticated-orcid":false,"given":"Xiaoge","family":"Deng","sequence":"additional","affiliation":[{"name":"Intelligent Game and Decision Lab, Beijing, China"}],"role":[{"role":"author","vocabulary":"crossref"}]},{"ORCID":"https:\/\/orcid.org\/0000-0002-6378-7002","authenticated-orcid":false,"given":"Kai","family":"Lu","sequence":"additional","affiliation":[{"name":"National Key Laboratory of Parallel and Distributed Computing, College of Computer Science and Technology, National University of Defense Technology, Changsha, Hunan, China"}],"role":[{"role":"author","vocabulary":"crossref"}]}],"member":"263","reference":[{"key":"ref1","article-title":"On the opportunities and risks of foundation models","author":"Bommasani","year":"2021"},{"key":"ref2","first-page":"1877","article-title":"Language models are few-shot learners","volume-title":"Proc. Int. Conf. Neural Inf. Process. Syst.","author":"Brown"},{"key":"ref3","article-title":"Scaling laws for neural language models","author":"Kaplan","year":"2020"},{"key":"ref4","article-title":"The cost of training NLP models: A concise overview","author":"Sharir","year":"2020"},{"key":"ref5","article-title":"Vast pricing","year":"2024"},{"key":"ref6","doi-asserted-by":"publisher","DOI":"10.14778\/3551793.3551828"},{"key":"ref7","doi-asserted-by":"publisher","DOI":"10.1145\/3575693.3575703"},{"key":"ref8","first-page":"381","article-title":"Fine-tuning giant neural networks on commodity hardware with automatic pipeline model parallelism","volume-title":"Proc. USENIX Annu. Tech. Conf.","author":"Eliad"},{"key":"ref9","article-title":"Megatron-LM: Training multi-billion parameter language models using model parallelism","author":"Shoeybi","year":"2019"},{"key":"ref10","first-page":"1","article-title":"Poseidon: An efficient communication architecture for distributed deep learning on GPU clusters","volume-title":"Proc. USENIX Annu. Tech. Conf.","author":"Zhang"},{"key":"ref11","doi-asserted-by":"publisher","DOI":"10.1109\/INFOCOM.2019.8737367"},{"key":"ref12","doi-asserted-by":"publisher","DOI":"10.1145\/3341301.3359642"},{"key":"ref13","first-page":"809","article-title":"Better together: Jointly optimizing ML collective scheduling and execution planning using SYNDICATE","volume-title":"Proc. 20th USENIX Symp. Netw. Syst. Des. Implementation","author":"Mahajan"},{"key":"ref14","doi-asserted-by":"publisher","DOI":"10.1145\/3567955.3567959"},{"key":"ref15","doi-asserted-by":"publisher","DOI":"10.1109\/TPDS.2022.3187815"},{"key":"ref16","doi-asserted-by":"publisher","DOI":"10.1145\/3503222.3507778"},{"key":"ref17","doi-asserted-by":"publisher","DOI":"10.1109\/TPDS.2023.3247001"},{"key":"ref18","first-page":"559","article-title":"Alpa: Automating inter-and intra-operator parallelism for distributed deep learning","volume-title":"Proc. 16th USENIX Symp. Operating Syst. Des. Implementation","author":"Zheng"},{"key":"ref19","article-title":"AMP: Automatically finding model parallel strategies with heterogeneity awareness","volume-title":"Proc. Int. Conf. Neural Inf. Process. Syst.","author":"Li"},{"key":"ref20","first-page":"267","article-title":"Unity: Accelerating DNN training through joint optimization of algebraic transformations and parallelization","volume-title":"Proc. 16th USENIX Symp. Operating Syst. Des. Implementation","author":"Unger"},{"key":"ref21","doi-asserted-by":"publisher","DOI":"10.14778\/3570690.3570697"},{"key":"ref22","article-title":"LLaMA: Open and efficient foundation language models","author":"Touvron","year":"2023"},{"key":"ref23","article-title":"OPT: Open pre-trained transformer language models","author":"Zhang","year":"2022"},{"key":"ref24","article-title":"BLOOM: A 176b-parameter open-access multilingual language model","author":"Scao","year":"2022"},{"key":"ref25","doi-asserted-by":"publisher","DOI":"10.48550\/ARXIV.1706.03762"},{"key":"ref26","doi-asserted-by":"publisher","DOI":"10.21236\/ADA561679"},{"key":"ref27","doi-asserted-by":"publisher","DOI":"10.1002\/(SICI)1096-9128(199704)9:4<255::AID-CPE250>3.0.CO;2-2"},{"key":"ref28","first-page":"673","article-title":"Whale: Efficient giant model training over heterogeneous GPUs","volume-title":"Proc. 2022 USENIX Annu. Tech. Conf.","author":"Jia"},{"key":"ref29","article-title":"End-to-end adaptive distributed training on paddlepaddle","author":"Ao","year":"2021"},{"key":"ref30","doi-asserted-by":"publisher","DOI":"10.1145\/3605573.3605613"},{"key":"ref31","first-page":"497","article-title":"Checkmate: Breaking the memory wall with optimal tensor rematerialization","volume-title":"Proc. 3rd Conf. Mach. Learn. Syst.","author":"Jain"},{"key":"ref32","article-title":"Training deep nets with sublinear memory cost","author":"Chen","year":"2016"},{"key":"ref33","article-title":"Dynamic tensor rematerialization","volume-title":"Proc. Int. Conf. Learn. Representations","author":"Kirisame"},{"key":"ref34","doi-asserted-by":"publisher","DOI":"10.1109\/TPDS.2023.3281931"},{"key":"ref35","article-title":"PanGu-$\\Sigma$\u03a3: Towards trillion parameter language model with sparse heterogeneous computing","author":"Ren","year":"2023"},{"key":"ref36","article-title":"Using DeepSpeed and megatron to train Megatron-turing NLG 530B, a large-scale generative language model","author":"Smith","year":"2022"},{"key":"ref37","article-title":"PyTorch: An imperative style, high-performance deep learning library","volume-title":"Proc. Int. Conf. Neural Inf. Process. Syst.","author":"Paszke"},{"key":"ref38","first-page":"1","article-title":"Efficient large-scale language model training on GPU clusters using megatron-LM","volume-title":"Proc. Int. Conf. High Perform. Comput. Netw. Storage Anal.","author":"Narayanan"},{"key":"ref39","doi-asserted-by":"publisher","DOI":"10.1016\/j.jpdc.2008.09.002"},{"key":"ref40","doi-asserted-by":"publisher","DOI":"10.1177\/1094342005051521"},{"key":"ref41","doi-asserted-by":"publisher","DOI":"10.1287\/educ.1053.0020"},{"key":"ref42","article-title":"BERT: Pre-training of deep bidirectional transformers for language understanding","author":"Devlin","year":"2018"},{"key":"ref43","doi-asserted-by":"publisher","DOI":"10.18653\/v1\/2022.bigscience-1.9"},{"key":"ref44","doi-asserted-by":"publisher","DOI":"10.1109\/TPDS.2019.2928289"},{"key":"ref45","doi-asserted-by":"publisher","DOI":"10.48550\/arxiv.1811.06965"},{"key":"ref46","first-page":"341","article-title":"Reducing activation recomputation in large transformer models","volume-title":"Proc. Int. Conf. Mach. Learn. Syst.","author":"Korthikanti"},{"key":"ref47","volume-title":"CUDA by Example: An Introduction to General-Purpose GPU Programming","author":"Sanders","year":"2010"},{"key":"ref48","doi-asserted-by":"publisher","DOI":"10.14778\/3415478.3415530"},{"key":"ref49","article-title":"Horovod: Fast and easy distributed deep learning in tensorflow","author":"Sergeev","year":"2018"},{"key":"ref50","doi-asserted-by":"publisher","DOI":"10.1109\/SC41404.2022.00076"},{"key":"ref51","doi-asserted-by":"publisher","DOI":"10.1109\/SC41405.2020.00024"},{"key":"ref52","doi-asserted-by":"publisher","DOI":"10.14778\/3611540.3611569"},{"key":"ref53","doi-asserted-by":"publisher","DOI":"10.1109\/HPCA56546.2023.10071077"},{"key":"ref54","doi-asserted-by":"publisher","DOI":"10.1109\/CLUSTER51413.2022.00042"},{"key":"ref55","doi-asserted-by":"publisher","DOI":"10.1145\/3545008.3545087"},{"key":"ref56","article-title":"Amazon SageMaker model parallelism: A general and flexible framework for large model training","author":"Karakus","year":"2021"},{"key":"ref57","doi-asserted-by":"publisher","DOI":"10.1145\/3437801.3441593"},{"key":"ref58","doi-asserted-by":"publisher","DOI":"10.1145\/3341301.3359646"},{"key":"ref59","first-page":"6543","article-title":"TeraPipe: Token-level pipeline parallelism for training large-scale language models","volume-title":"Proc. Int. Conf. Mach. Learn.","author":"Li"},{"key":"ref60","first-page":"1","article-title":"Beyond data and model parallelism for deep neural networks","volume-title":"Proc. Int. Conf. Mach. Learn. Syst.","author":"Jia"},{"key":"ref61","first-page":"24829","article-title":"Piper: Multidimensional planner for DNN parallelization","volume-title":"Proc. Int. Conf. Neural Inf. Process. Syst.","author":"Tarnawski"},{"key":"ref62","doi-asserted-by":"publisher","DOI":"10.1145\/3545008.3545011"},{"key":"ref63","doi-asserted-by":"publisher","DOI":"10.1145\/3492321.3519563"},{"key":"ref64","doi-asserted-by":"publisher","DOI":"10.1145\/3472456.3472497"},{"key":"ref65","first-page":"526","article-title":"On optimizing the communication of model parallelism","volume-title":"Proc. Int. Conf. Mach. Learn. Syst.","author":"Zhuang"}],"container-title":["IEEE Transactions on Parallel and Distributed Systems"],"original-title":[],"link":[{"URL":"http:\/\/xplorestaging.ieee.org\/ielx8\/71\/11083525\/11050994.pdf?arnumber=11050994","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2025,7,18]],"date-time":"2025-07-18T17:47:05Z","timestamp":1752860825000},"score":1,"resource":{"primary":{"URL":"https:\/\/ieeexplore.ieee.org\/document\/11050994\/"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2025,9]]},"references-count":65,"journal-issue":{"issue":"9"},"URL":"https:\/\/doi.org\/10.1109\/tpds.2025.3583165","relation":{},"ISSN":["1045-9219","1558-2183","2161-9883"],"issn-type":[{"value":"1045-9219","type":"print"},{"value":"1558-2183","type":"electronic"},{"value":"2161-9883","type":"electronic"}],"subject":[],"published":{"date-parts":[[2025,9]]}}}