{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2025,12,18]],"date-time":"2025-12-18T12:31:46Z","timestamp":1766061106915,"version":"3.48.0"},"reference-count":46,"publisher":"Institute of Electrical and Electronics Engineers (IEEE)","license":[{"start":{"date-parts":[[2025,1,1]],"date-time":"2025-01-01T00:00:00Z","timestamp":1735689600000},"content-version":"vor","delay-in-days":0,"URL":"https:\/\/creativecommons.org\/licenses\/by\/4.0\/legalcode"}],"funder":[{"DOI":"10.13039\/100004689","name":"Booz Allen Hamilton","doi-asserted-by":"publisher","id":[{"id":"10.13039\/100004689","id-type":"DOI","asserted-by":"publisher"}]}],"content-domain":{"domain":[],"crossmark-restriction":false},"short-container-title":["IEEE Access"],"published-print":{"date-parts":[[2025]]},"DOI":"10.1109\/access.2025.3625771","type":"journal-article","created":{"date-parts":[[2025,10,27]],"date-time":"2025-10-27T18:05:43Z","timestamp":1761588343000},"page":"209356-209367","source":"Crossref","is-referenced-by-count":0,"title":["Sub 4-bit Power-of-Two-Based Mixed-Precision Quantization for Efficient LLM Compression and Acceleration"],"prefix":"10.1109","volume":"13","author":[{"ORCID":"https:\/\/orcid.org\/0009-0000-9010-7888","authenticated-orcid":false,"given":"Han","family":"Cho","sequence":"first","affiliation":[{"name":"School of Electrical and Computer Engineering, Georgia Institute of Technology, Atlanta, GA, USA"}],"role":[{"role":"author","vocabulary":"crossref"}]},{"ORCID":"https:\/\/orcid.org\/0009-0008-0401-3809","authenticated-orcid":false,"given":"Apurba","family":"Prasad Padhy","sequence":"additional","affiliation":[{"name":"School of Electrical and Computer Engineering, Georgia Institute of Technology, Atlanta, GA, USA"}],"role":[{"role":"author","vocabulary":"crossref"}]},{"ORCID":"https:\/\/orcid.org\/0009-0006-3181-4076","authenticated-orcid":false,"given":"Fernando","family":"Camacho","sequence":"additional","affiliation":[{"name":"Laboratory for Physical Sciences, College Park, MD, USA"}],"role":[{"role":"author","vocabulary":"crossref"}]},{"ORCID":"https:\/\/orcid.org\/0000-0002-8894-3390","authenticated-orcid":false,"given":"Saibal","family":"Mukhopadhyay","sequence":"additional","affiliation":[{"name":"School of Electrical and Computer Engineering, Georgia Institute of Technology, Atlanta, GA, USA"}],"role":[{"role":"author","vocabulary":"crossref"}]}],"member":"263","reference":[{"key":"ref1","first-page":"1877","article-title":"Language models are few-shot learners","volume-title":"Proc. Adv. Neural Inf. Process. Syst.","author":"Brown"},{"key":"ref2","article-title":"LLaMA: Open and efficient foundation language models","author":"Touvron","year":"2023","journal-title":"arXiv:2302.13971"},{"key":"ref3","article-title":"OPT: Open pre-trained transformer language models","author":"Zhang","year":"2022","journal-title":"arXiv:2205.01068"},{"key":"ref4","article-title":"BERT: Pre-training of deep bidirectional transformers for language understanding","author":"Devlin","year":"2018","journal-title":"arXiv:1810.04805"},{"key":"ref5","article-title":"Llama 2: Open foundation and fine-tuned chat models","author":"Touvron","year":"2023","journal-title":"arXiv:2307.09288"},{"key":"ref6","article-title":"Think you have solved question answering? Try ARC, the AI2 reasoning challenge","author":"Clark","year":"2018","journal-title":"arXiv:1803.05457"},{"key":"ref7","article-title":"Textbooks are all you need","author":"Gunasekar","year":"2023","journal-title":"arXiv:2306.11644"},{"key":"ref8","article-title":"Apple intelligence foundation language models","author":"Gunter","year":"2024","journal-title":"arXiv:2407.21075"},{"key":"ref9","doi-asserted-by":"publisher","DOI":"10.1109\/OJCOMS.2024.3456549"},{"key":"ref10","first-page":"30318","article-title":"Gpt3. Int8 (): 8-bit matrix multiplication for transformers at scale","volume-title":"Proc. Adv. Neural Inf. Process. Syst.","volume":"35","author":"Dettmers"},{"key":"ref11","first-page":"1","article-title":"LUT-GEMM: Quantized matrix multiplication based on LUTs for efficient inference in large-scale generative language models","volume-title":"Proc. Int. Conf. Learn. Represent.","author":"Park"},{"key":"ref12","article-title":"GPTQ: Accurate post-training quantization for generative pre-trained transformers","author":"Frantar","year":"2022","journal-title":"arXiv:2210.17323"},{"key":"ref13","first-page":"12284","article-title":"Extreme compression of large language models via additive quantization","volume-title":"Proc. Int. Conf. Mach. Learn.","author":"Egiazarian"},{"key":"ref14","first-page":"7197","article-title":"Up or down? Adaptive rounding for post-training quantization","volume-title":"Proc. Int. Conf. Mach. Learn.","author":"Nagel"},{"key":"ref15","first-page":"1","article-title":"Deep compression: Compressing deep neural networks with pruning, trained quantization and Huffman coding","volume-title":"Proc. Int. Conf. Learn. Represent.","author":"Han"},{"key":"ref16","article-title":"I-LLM: Efficient integer-only inference for fully-quantized low-bit large language models","author":"Hu","year":"2024","journal-title":"arXiv:2405.17849"},{"key":"ref17","doi-asserted-by":"publisher","DOI":"10.18653\/v1\/2024.findings-acl.26"},{"key":"ref18","article-title":"EfficientQAT: Efficient quantization-aware training for large language models","author":"Chen","year":"2024","journal-title":"arXiv:2407.11062"},{"key":"ref19","first-page":"1","article-title":"BRECQ: Pushing the limit of post-training quantization by block reconstruction","volume-title":"Proc. Int. Conf. Learn. Represent.","author":"Li"},{"key":"ref20","doi-asserted-by":"publisher","DOI":"10.1109\/ICCV.2019.00141"},{"key":"ref21","doi-asserted-by":"publisher","DOI":"10.1109\/ICCV.2019.00038"},{"key":"ref22","first-page":"18518","article-title":"HAWQ-v2: Hessian aware trace-weighted quantization of neural networks","volume-title":"Proc. Adv. Neural Inf. Process. Syst.","author":"Dong"},{"key":"ref23","first-page":"4475","article-title":"Optimal brain compression: A framework for accurate post-training quantization and pruning","volume-title":"Proc. Adv. Neural Inf. Process. Syst.","author":"Frantar"},{"key":"ref24","doi-asserted-by":"publisher","DOI":"10.1609\/aaai.v38i12.29237"},{"key":"ref25","doi-asserted-by":"publisher","DOI":"10.1145\/3714983.3714987"},{"key":"ref26","first-page":"1","article-title":"OmniQuant: Omnidirectionally calibrated quantization for large language models","volume-title":"Proc. Int. Conf. Learn. Represent.","author":"Shao"},{"key":"ref27","doi-asserted-by":"publisher","DOI":"10.1609\/aaai.v38i17.29908"},{"key":"ref28","first-page":"1","article-title":"QDrop: Randomly dropping quantization for extremely low-bit post-training quantization","volume-title":"Proc. Int. Conf. Learn. Represent.","author":"Wei"},{"key":"ref29","doi-asserted-by":"publisher","DOI":"10.18653\/v1\/P19-1472"},{"key":"ref30","article-title":"Measuring massive multitask language understanding","author":"Hendrycks","year":"2020","journal-title":"arXiv:2009.03300"},{"key":"ref31","doi-asserted-by":"publisher","DOI":"10.1145\/3477002"},{"key":"ref32","doi-asserted-by":"publisher","DOI":"10.1145\/3370748.3406567"},{"key":"ref33","doi-asserted-by":"publisher","DOI":"10.1145\/3431920.3439477"},{"key":"ref34","doi-asserted-by":"publisher","DOI":"10.1109\/ICCAD51958.2021.9643586"},{"key":"ref35","doi-asserted-by":"publisher","DOI":"10.1109\/TPDS.2020.3047371"},{"key":"ref36","doi-asserted-by":"publisher","DOI":"10.1109\/ISQED51717.2021.9424344"},{"key":"ref37","doi-asserted-by":"publisher","DOI":"10.1109\/TCAD.2022.3197489"},{"key":"ref38","article-title":"VAQF: Fully automatic software-hardware co-design framework for low-bit vision transformer","author":"Sun","year":"2022","journal-title":"arXiv:2201.06618"},{"key":"ref39","doi-asserted-by":"publisher","DOI":"10.1109\/HiPC58850.2023.00039"},{"key":"ref40","doi-asserted-by":"publisher","DOI":"10.1109\/ASP-DAC58780.2024.10473931"},{"key":"ref41","doi-asserted-by":"publisher","DOI":"10.1145\/3676536.3676681"},{"key":"ref42","doi-asserted-by":"publisher","DOI":"10.1109\/ISCAS58744.2024.10557992"},{"key":"ref43","doi-asserted-by":"publisher","DOI":"10.1145\/3626202.3637569"},{"key":"ref44","doi-asserted-by":"publisher","DOI":"10.1145\/3626202.3637562"},{"key":"ref45","doi-asserted-by":"publisher","DOI":"10.23919\/DATE58400.2024.10546652"},{"key":"ref46","doi-asserted-by":"publisher","DOI":"10.1145\/3649476.3658810"}],"container-title":["IEEE Access"],"original-title":[],"link":[{"URL":"http:\/\/xplorestaging.ieee.org\/ielx8\/6287639\/10820123\/11218199.pdf?arnumber=11218199","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2025,12,18]],"date-time":"2025-12-18T12:29:01Z","timestamp":1766060941000},"score":1,"resource":{"primary":{"URL":"https:\/\/ieeexplore.ieee.org\/document\/11218199\/"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2025]]},"references-count":46,"URL":"https:\/\/doi.org\/10.1109\/access.2025.3625771","relation":{},"ISSN":["2169-3536"],"issn-type":[{"type":"electronic","value":"2169-3536"}],"subject":[],"published":{"date-parts":[[2025]]}}}