{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2025,12,19]],"date-time":"2025-12-19T07:08:25Z","timestamp":1766128105785,"version":"3.48.0"},"reference-count":59,"publisher":"IEEE","license":[{"start":{"date-parts":[[2025,11,3]],"date-time":"2025-11-03T00:00:00Z","timestamp":1762128000000},"content-version":"stm-asf","delay-in-days":0,"URL":"https:\/\/doi.org\/10.15223\/policy-029"},{"start":{"date-parts":[[2025,11,3]],"date-time":"2025-11-03T00:00:00Z","timestamp":1762128000000},"content-version":"stm-asf","delay-in-days":0,"URL":"https:\/\/doi.org\/10.15223\/policy-037"}],"funder":[{"DOI":"10.13039\/501100003725","name":"National Research Foundation of Korea","doi-asserted-by":"publisher","id":[{"id":"10.13039\/501100003725","id-type":"DOI","asserted-by":"publisher"}]},{"DOI":"10.13039\/100018696","name":"Health","doi-asserted-by":"publisher","id":[{"id":"10.13039\/100018696","id-type":"DOI","asserted-by":"publisher"}]},{"DOI":"10.13039\/100018696","name":"Health","doi-asserted-by":"publisher","id":[{"id":"10.13039\/100018696","id-type":"DOI","asserted-by":"publisher"}]},{"DOI":"10.13039\/501100010238","name":"K2","doi-asserted-by":"publisher","id":[{"id":"10.13039\/501100010238","id-type":"DOI","asserted-by":"publisher"}]},{"DOI":"10.13039\/501100002551","name":"Seoul National University","doi-asserted-by":"publisher","id":[{"id":"10.13039\/501100002551","id-type":"DOI","asserted-by":"publisher"}]},{"DOI":"10.13039\/501100002701","name":"Ministry of Education","doi-asserted-by":"publisher","id":[{"id":"10.13039\/501100002701","id-type":"DOI","asserted-by":"publisher"}]}],"content-domain":{"domain":[],"crossmark-restriction":false},"short-container-title":[],"published-print":{"date-parts":[[2025,11,3]]},"DOI":"10.1109\/pact65351.2025.00014","type":"proceedings-article","created":{"date-parts":[[2025,12,16]],"date-time":"2025-12-16T18:30:30Z","timestamp":1765909830000},"page":"14-29","source":"Crossref","is-referenced-by-count":0,"title":["SPipe: Hybrid GPU and CPU Pipeline for Training LLMs under Memory Pressure"],"prefix":"10.1109","author":[{"given":"Junyeol","family":"Ryu","sequence":"first","affiliation":[{"name":"Seoul National University,Dept. of Computer Science and Engineering,Seoul,South Korea"}]},{"given":"Yujin","family":"Jeong","sequence":"additional","affiliation":[{"name":"Seoul National University,Dept. of Computer Science and Engineering,Seoul,South Korea"}]},{"given":"Daeyoung","family":"Park","sequence":"additional","affiliation":[{"name":"Seoul National University,Dept. of Computer Science and Engineering,Seoul,South Korea"}]},{"given":"Jinpyo","family":"Kim","sequence":"additional","affiliation":[{"name":"Seoul National University,Dept. of Computer Science and Engineering,Seoul,South Korea"}]},{"given":"Heehoon","family":"Kim","sequence":"additional","affiliation":[{"name":"Seoul National University,Dept. of Computer Science and Engineering,Seoul,South Korea"}]},{"given":"Jaejin","family":"Lee","sequence":"additional","affiliation":[{"name":"Seoul National University,Dept. of Computer Science and Engineering,Seoul,South Korea"}]}],"member":"263","reference":[{"key":"ref1","doi-asserted-by":"publisher","DOI":"10.1145\/3492321.3519584"},{"key":"ref2","first-page":"387","article-title":"FlashNeuron: SSD-enabled large-batch training of very deep neural networks","volume-title":"19th USENIX Conference on File and Storage Technologies (FAST 21)","author":"Bae"},{"key":"ref3","doi-asserted-by":"publisher","DOI":"10.1145\/3320060"},{"key":"ref4","article-title":"Language models are few-shot learners","author":"Brown","year":"2020","journal-title":"arXiv preprint arXiv:2005.14165"},{"year":"2024","key":"ref5","article-title":"std::thread"},{"key":"ref6","article-title":"Training deep nets with sublinear memory cost","author":"Chen","year":"2016","journal-title":"arXiv preprint arXiv:1604.06174"},{"key":"ref7","doi-asserted-by":"publisher","DOI":"10.1109\/TPDS.2020.3048836"},{"key":"ref8","doi-asserted-by":"publisher","DOI":"10.1145\/3437801.3441593"},{"key":"ref9","doi-asserted-by":"publisher","DOI":"10.1145\/3575693.3575703"},{"key":"ref10","doi-asserted-by":"publisher","DOI":"10.1145\/3694715.3695960"},{"key":"ref11","doi-asserted-by":"publisher","DOI":"10.1145\/3373376.3378465"},{"key":"ref12","doi-asserted-by":"publisher","DOI":"10.1145\/3373376.3378530"},{"key":"ref13","doi-asserted-by":"publisher","DOI":"10.48550\/arxiv.1811.06965"},{"key":"ref14","first-page":"497","article-title":"Checkmate: Breaking the memory wall with optimal tensor rematerialization","volume":"2","author":"Jain","year":"2020","journal-title":"Proceedings of Machine Learning and Systems"},{"key":"ref15","doi-asserted-by":"publisher","DOI":"10.1145\/3243904"},{"key":"ref16","article-title":"Scaling laws for neural language models","author":"Kaplan","year":"2020","journal-title":"arXiv preprint arXiv:2001.08361"},{"key":"ref17","article-title":"Adam: A method for stochastic optimization","author":"Kingma","year":"2014","journal-title":"arXiv preprint arXiv:1412.6980"},{"article-title":"Dynamic tensor rematerialization","volume-title":"International Conference on Learning Representations","author":"Kirisame","key":"ref18"},{"key":"ref19","article-title":"Reducing activation recomputation in large transformer models","volume":"5","author":"Korthikanti","year":"2023","journal-title":"Proceedings of Machine Learning and Systems"},{"key":"ref20","article-title":"TFLMS: Large model support in TensorFlow by graph rewriting","author":"Le","year":"2018","journal-title":"arXiv preprint arXiv:1807.02037"},{"key":"ref21","doi-asserted-by":"publisher","DOI":"10.1145\/3458817.3476145"},{"key":"ref22","doi-asserted-by":"publisher","DOI":"10.1145\/3605573.3605647"},{"key":"ref23","doi-asserted-by":"publisher","DOI":"10.1145\/3178487.3178528"},{"key":"ref24","doi-asserted-by":"publisher","DOI":"10.1145\/3581784.3607073"},{"article-title":"Decoupled weight decay regularization","volume-title":"International Conference on Learning Representations","author":"Loshchilov","key":"ref25"},{"key":"ref26","doi-asserted-by":"publisher","DOI":"10.1145\/3659995.3660038"},{"key":"ref27","doi-asserted-by":"publisher","DOI":"10.1145\/3652892.3700781"},{"year":"2024","key":"ref28","article-title":"LLaMA3"},{"key":"ref29","article-title":"Mixed precision training","author":"Micikevicius","year":"2017","journal-title":"arXiv preprint arXiv:1710.03740"},{"year":"2024","key":"ref30","article-title":"DeepSpeed"},{"key":"ref31","doi-asserted-by":"publisher","DOI":"10.1145\/3341301.3359646"},{"key":"ref32","first-page":"7937","article-title":"Memory-efficient pipeline-parallel DNN training","volume-title":"International Conference on Machine Learning.","author":"Narayanan"},{"key":"ref33","first-page":"1","article-title":"Efficient large-scale language model training on GPU clusters using Megatron-LM","volume-title":"Proceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis,","author":"Narayanan"},{"year":"2024","key":"ref34","article-title":"Megatron-LM"},{"key":"ref35","doi-asserted-by":"publisher","DOI":"10.1145\/3373376.3378505"},{"article-title":"OpenWebText","year":"2019","author":"Peterson","key":"ref36"},{"year":"2024","key":"ref37","article-title":"The Open Group Base Specifications"},{"key":"ref38","article-title":"Training large neural networks with constant memory using a new execution algorithm","author":"Pudipeddi","year":"2020","journal-title":"arXiv preprint arXiv:2002.05645"},{"key":"ref39","article-title":"Zero bubble pipeline parallelism","author":"Qi","year":"2023","journal-title":"arXiv preprint arXiv:2401.10241"},{"article-title":"Zero bubble (almost) pipeline parallelism","volume-title":"The Twelfth International Conference on Learning Representations","author":"Qi","key":"ref40"},{"issue":"8","key":"ref41","first-page":"9","article-title":"Language models are unsupervised multitask learners","volume":"1","author":"Radford","year":"2019","journal-title":"OpenAI blog"},{"key":"ref42","doi-asserted-by":"publisher","DOI":"10.1109\/SC41405.2020.00024"},{"key":"ref43","first-page":"1","article-title":"ZeROInfinity: Breaking the GPU memory wall for extreme scale deep learning","volume-title":"Proceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis","author":"Rajbhandari"},{"key":"ref44","doi-asserted-by":"publisher","DOI":"10.1145\/3394486.3406703"},{"key":"ref45","doi-asserted-by":"publisher","DOI":"10.1109\/HPCA51647.2021.00057"},{"key":"ref46","first-page":"551","article-title":"ZeRO-Offload: Democratizing billion-scale model training","volume-title":"2021 USENIX Annual Technical Conference (USENIX ATC 21),","author":"Ren"},{"key":"ref47","doi-asserted-by":"publisher","DOI":"10.1109\/MICRO.2016.7783721"},{"key":"ref48","doi-asserted-by":"publisher","DOI":"10.1145\/3624724"},{"key":"ref49","article-title":"Megatron-LM: Training multi-billion parameter language models using model parallelism","author":"Shoeybi","year":"2019","journal-title":"arXiv preprint arXiv:1909.08053"},{"key":"ref50","article-title":"LLaMA: Open and efficient foundation language models","author":"Touvron","year":"2023","journal-title":"arXiv preprint arXiv:2302.13971"},{"key":"ref51","article-title":"LLaMA 2: Open foundation and fine-tuned chat models","author":"Touvron","year":"2023","journal-title":"arXiv preprint arXiv:2307.09288"},{"article-title":"Zero++: Extremely efficient collective communication for large model training","volume-title":"The Twelfth International Conference on Learning Representations","author":"Wang","key":"ref52"},{"key":"ref53","first-page":"1","article-title":"Large-batch training for LSTM and beyond","volume-title":"Proceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis","author":"You"},{"key":"ref54","article-title":"Large batch optimization for deep learning: Training BERT in 76 minutes","author":"You","year":"2019","journal-title":"arXiv preprint arXiv:1904.00962"},{"key":"ref55","doi-asserted-by":"publisher","DOI":"10.1145\/3225058.3225069"},{"key":"ref56","article-title":"OPT: Open pre-trained transformer language models","author":"Zhang","year":"2022","journal-title":"arXiv preprint arXiv:2205.01068"},{"key":"ref57","doi-asserted-by":"publisher","DOI":"10.1109\/DAC56929.2023.10247730"},{"key":"ref58","article-title":"A survey of large language models","author":"Zhao","year":"2023","journal-title":"arXiv preprint arXiv:2303.18223"},{"key":"ref59","first-page":"559","article-title":"Alpa: Automating inter-and intra-operator parallelism for distributed deep learning","volume-title":"16th USENIX Symposium on Operating Systems Design and Implementation (OSDI 22)","author":"Zheng"}],"event":{"name":"2025 34th International Conference on Parallel Architectures and Compilation Techniques (PACT)","start":{"date-parts":[[2025,11,3]]},"location":"Irvine, CA, USA","end":{"date-parts":[[2025,11,6]]}},"container-title":["2025 34th International Conference on Parallel Architectures and Compilation Techniques (PACT)"],"original-title":[],"link":[{"URL":"http:\/\/xplorestaging.ieee.org\/ielx8\/11282903\/11282500\/11282926.pdf?arnumber=11282926","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2025,12,19]],"date-time":"2025-12-19T07:03:29Z","timestamp":1766127809000},"score":1,"resource":{"primary":{"URL":"https:\/\/ieeexplore.ieee.org\/document\/11282926\/"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2025,11,3]]},"references-count":59,"URL":"https:\/\/doi.org\/10.1109\/pact65351.2025.00014","relation":{},"subject":[],"published":{"date-parts":[[2025,11,3]]}}}