{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2025,10,6]],"date-time":"2025-10-06T18:11:19Z","timestamp":1759774279843,"version":"build-2065373602"},"reference-count":61,"publisher":"Institute of Electrical and Electronics Engineers (IEEE)","issue":"11","license":[{"start":{"date-parts":[[2025,11,1]],"date-time":"2025-11-01T00:00:00Z","timestamp":1761955200000},"content-version":"vor","delay-in-days":0,"URL":"https:\/\/ieeexplore.ieee.org\/Xplorehelp\/downloads\/license-information\/IEEE.html"},{"start":{"date-parts":[[2025,11,1]],"date-time":"2025-11-01T00:00:00Z","timestamp":1761955200000},"content-version":"stm-asf","delay-in-days":0,"URL":"https:\/\/doi.org\/10.15223\/policy-029"},{"start":{"date-parts":[[2025,11,1]],"date-time":"2025-11-01T00:00:00Z","timestamp":1761955200000},"content-version":"stm-asf","delay-in-days":0,"URL":"https:\/\/doi.org\/10.15223\/policy-037"}],"funder":[{"DOI":"10.13039\/501100003032","name":"Association Nationale de la Recherche et de la Technologie","doi-asserted-by":"publisher","id":[{"id":"10.13039\/501100003032","id-type":"DOI","asserted-by":"publisher"}]},{"name":"French National Agency"},{"name":"HPC resources of IDRIS"}],"content-domain":{"domain":[],"crossmark-restriction":false},"short-container-title":["IEEE Trans. Pattern Anal. Mach. Intell."],"published-print":{"date-parts":[[2025,11]]},"DOI":"10.1109\/tpami.2025.3593987","type":"journal-article","created":{"date-parts":[[2025,8,4]],"date-time":"2025-08-04T18:45:09Z","timestamp":1754333109000},"page":"10012-10021","source":"Crossref","is-referenced-by-count":0,"title":["NUPES: Non-Uniform Post-Training Quantization via Power Exponent Search"],"prefix":"10.1109","volume":"47","author":[{"ORCID":"https:\/\/orcid.org\/0000-0002-4318-612X","authenticated-orcid":false,"given":"Edouard","family":"Yvinec","sequence":"first","affiliation":[{"name":"Datakalab, Paris, France"}],"role":[{"role":"author","vocabulary":"crossref"}]},{"ORCID":"https:\/\/orcid.org\/0000-0002-0074-8719","authenticated-orcid":false,"given":"Arnaud","family":"Dapogny","sequence":"additional","affiliation":[{"name":"Datakalab, Paris, France"}],"role":[{"role":"author","vocabulary":"crossref"}]},{"ORCID":"https:\/\/orcid.org\/0000-0001-7802-3673","authenticated-orcid":false,"given":"Kevin","family":"Bailly","sequence":"additional","affiliation":[{"name":"Datakalab, Paris, France"}],"role":[{"role":"author","vocabulary":"crossref"}]}],"member":"263","reference":[{"key":"ref1","first-page":"265","article-title":"TensorFlow: Large-scale machine learning on heterogeneous systems","volume-title":"Proc. 12th USENIX Conf. Operating Syst. Des. Implementation","author":"Abadi","year":"2015"},{"key":"ref2","first-page":"7950","article-title":"Post training 4-bit quantization of convolutional networks for rapid-deployment","volume-title":"Proc. Int. Conf. Neural Inf. Process. Syst.","author":"Banner","year":"2019"},{"key":"ref3","doi-asserted-by":"publisher","DOI":"10.1609\/aaai.v34i05.6239"},{"key":"ref4","first-page":"2285","article-title":"Compressing neural networks with the hashing trick","volume-title":"Proc. Int. Conf. Mach. Learn.","author":"Chen","year":"2015"},{"key":"ref5","doi-asserted-by":"publisher","DOI":"10.1109\/msp.2017.2765695"},{"article-title":"Exploring the surprising difficulty of natural yes\/no questions","year":"2019","author":"Clark","key":"ref6"},{"article-title":"Think you have solved question answering? Try arc, the AI2 reasoning challenge.","year":"2018","author":"Clark","key":"ref7"},{"key":"ref8","article-title":"Squant: On-the-fly data-free quantization via diagonal hessian approximation","volume-title":"Proc. Int. Conf. Learn. Representations","author":"Cong","year":"2022"},{"key":"ref9","first-page":"3123","article-title":"Binaryconnect: Training deep neural networks with binary weights during propagations","volume-title":"Proc. Int. Conf. Neural Inf. Process. Syst.","author":"Courbariaux","year":"2015"},{"key":"ref10","article-title":"Binarized neural networks: Training deep neural networks with weights and activations constrained to 1 or-1","volume-title":"Proc. Int. Conf. Neural Inf. Process. Syst.","author":"Courbariaux","year":"2016"},{"year":"2023","key":"ref11","article-title":"Databricks\u2019 dolly-V2, an instruction-following large language model trained on the databricks machine learning platform"},{"key":"ref12","doi-asserted-by":"publisher","DOI":"10.1109\/CVPR.2009.5206848"},{"article-title":"LLM. INT8(): 8-bit matrix multiplication for transformers at scale","year":"2022","author":"Dettmers","key":"ref13"},{"key":"ref14","first-page":"10088","article-title":"QLoRA: Efficient finetuning of quantized LLMs","volume-title":"Proc. Adv. Neural Inf. Process. Syst.","author":"Dettmers","year":"2023"},{"key":"ref15","article-title":"An image is worth 16x16 words: Transformers for image recognition at scale","volume-title":"Proc. Int. Conf. Learn. Representations","author":"Dosovitskiy","year":"2021"},{"article-title":"Torchvision: Models, datasets and transformations for images","year":"2023","author":"Falbel","key":"ref16"},{"key":"ref17","article-title":"OPTQ: Accurate quantization for generative pre-trained transformers","volume-title":"Proc. Int. Conf. Learn. Representations","author":"Frantar","year":"2022"},{"key":"ref18","article-title":"OPTQ: Accurate quantization for generative pre-trained transformers","volume-title":"Proc. Int. Conf. Learn. Representations","author":"Frantar","year":"2023"},{"key":"ref19","doi-asserted-by":"publisher","DOI":"10.1109\/ICCV.2019.00495"},{"article-title":"Compressing deep convolutional networks using vector quantization","year":"2014","author":"Gong","key":"ref20"},{"key":"ref21","doi-asserted-by":"publisher","DOI":"10.1109\/CVPR52688.2022.00978"},{"key":"ref22","doi-asserted-by":"publisher","DOI":"10.1109\/CVPR.2016.90"},{"key":"ref23","volume-title":"Axiom of Choice","volume":"1876","author":"Herrlich","year":"2006"},{"key":"ref24","first-page":"4114","article-title":"Binarized neural networks","volume-title":"Proc. Int. Conf. Neural Inf. Process. Syst.","author":"Hubara","year":"2016"},{"year":"2023","key":"ref25","article-title":"The opt model was proposed in open pre-trained transformer language models by meta AI"},{"key":"ref26","doi-asserted-by":"publisher","DOI":"10.1109\/SC41405.2020.00099"},{"article-title":"Adam: A method for stochastic optimization","year":"2014","author":"Kingma","key":"ref27"},{"article-title":"Quantizing deep convolutional networks for efficient inference: A whitepaper","year":"2018","author":"Krishnamoorthi","key":"ref28"},{"key":"ref29","article-title":"Brecq: Pushing the limit of post-training quantization by block reconstruction","volume-title":"Proc. Int. Conf. Learn. Representations","author":"Li","year":"2021"},{"key":"ref30","doi-asserted-by":"publisher","DOI":"10.1007\/978-3-319-10602-1_48"},{"key":"ref31","doi-asserted-by":"publisher","DOI":"10.1109\/ICCV.2017.324"},{"key":"ref32","doi-asserted-by":"publisher","DOI":"10.1109\/CVPR52729.2023.02340"},{"article-title":"Pointersentinelmixture models","year":"2016","author":"Merity","key":"ref33"},{"key":"ref34","doi-asserted-by":"publisher","DOI":"10.18653\/v1\/D18-1260"},{"article-title":"Convolutional neural networks using logarithmic data representation","year":"2016","author":"Miyashita","key":"ref35"},{"key":"ref36","first-page":"7197","article-title":"Up or down? Adaptive rounding for post-training quantization","volume-title":"Proc. Int. Conf. Mach. Learn.","author":"Nagel","year":"2020"},{"key":"ref37","doi-asserted-by":"publisher","DOI":"10.1109\/ICCV.2019.00141"},{"key":"ref38","doi-asserted-by":"publisher","DOI":"10.1093\/comjnl\/7.4.308"},{"article-title":"Nuqmm: Quantized matmul for efficient inference of large-scale generative language models","year":"2022","author":"Park","key":"ref39"},{"year":"2022","key":"ref40","article-title":"Tai model efficiency toolkit (aimet)"},{"article-title":"An overview of gradient descent optimization algorithms","year":"2016","author":"Ruder","key":"ref41"},{"key":"ref42","doi-asserted-by":"publisher","DOI":"10.1145\/3474381"},{"key":"ref43","doi-asserted-by":"publisher","DOI":"10.1109\/CVPR.2018.00474"},{"key":"ref44","article-title":"Omniquant: Omnidirectionally calibrated quantization for large language models","volume-title":"Proc. Int. Conf. Learn. Representations","author":"Shao","year":"2024"},{"key":"ref45","first-page":"6105","article-title":"EfficientNet: Rethinking model scaling for convolutional neural networks","volume-title":"Proc. Int. Conf. Mach. Learn.","author":"Tan","year":"2019"},{"year":"2025","key":"ref46","article-title":"Gemma 3 technical report"},{"article-title":"The llama 4 herd: The beginning of a new era of natively multimodal ai innovation","year":"2025","author":"Team","key":"ref47"},{"key":"ref48","article-title":"Quip#: Even better LLM quantization with hadamard incoherence and lattice codebooks","volume-title":"Proc. Int. Conf. Learn. Representations","author":"Tseng","year":"2024"},{"key":"ref49","article-title":"Attention is all you need","volume-title":"Proc. Int. Conf. Neural Inf. Process. Syst.","author":"Vaswani","year":"2017"},{"key":"ref50","article-title":"Qdrop: Randomly dropping quantization for extremely low-bit post-training quantization","volume-title":"Proc. Int. Conf. Learn. Representations","author":"Wei","year":"2022"},{"key":"ref51","first-page":"38087","article-title":"Smoothquant: Accurate and efficient post-training quantization for large language models","volume-title":"Proc. Int. Conf. Mach. Learn.","author":"Xiao","year":"2023"},{"key":"ref52","doi-asserted-by":"publisher","DOI":"10.1007\/978-3-030-58610-2_1"},{"key":"ref53","first-page":"44925","article-title":"REX: Data-free residual quantization error expansion","volume-title":"Proc. Int. Conf. Neural Inf. Process. Syst.","author":"Yvinec","year":"2023"},{"key":"ref54","article-title":"Powerquant: Automorphism search for non-uniform quantization","volume-title":"Proc. Int. Conf. Learn. Representations","author":"Yvinec","year":"2023"},{"key":"ref55","doi-asserted-by":"publisher","DOI":"10.1109\/WACV56688.2023.00386"},{"key":"ref56","doi-asserted-by":"publisher","DOI":"10.18653\/v1\/P19-1472"},{"article-title":"Open pre-trained transformer language models","year":"2022","author":"Zhang","key":"ref57"},{"key":"ref58","doi-asserted-by":"publisher","DOI":"10.1145\/3445814.3446741"},{"key":"ref59","doi-asserted-by":"publisher","DOI":"10.1109\/CVPR52688.2022.01215"},{"key":"ref60","first-page":"7543","article-title":"Improving neural network quantization without retraining using outlier channel splitting","volume-title":"Proc. Int. Conf. Mach. Learn.","author":"Zhao","year":"2019"},{"key":"ref61","article-title":"Incremental network quantization: Towards lossless CNNs with low-precision weights","volume-title":"Proc. Int. Conf. Learn. Representations","author":"Zhou","year":"2017"}],"container-title":["IEEE Transactions on Pattern Analysis and Machine Intelligence"],"original-title":[],"link":[{"URL":"http:\/\/xplorestaging.ieee.org\/ielx8\/34\/11192800\/11112770.pdf?arnumber=11112770","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2025,10,6]],"date-time":"2025-10-06T17:36:40Z","timestamp":1759772200000},"score":1,"resource":{"primary":{"URL":"https:\/\/ieeexplore.ieee.org\/document\/11112770\/"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2025,11]]},"references-count":61,"journal-issue":{"issue":"11"},"URL":"https:\/\/doi.org\/10.1109\/tpami.2025.3593987","relation":{},"ISSN":["0162-8828","2160-9292","1939-3539"],"issn-type":[{"type":"print","value":"0162-8828"},{"type":"electronic","value":"2160-9292"},{"type":"electronic","value":"1939-3539"}],"subject":[],"published":{"date-parts":[[2025,11]]}}}