{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,6,30]],"date-time":"2026-06-30T05:45:37Z","timestamp":1782798337228,"version":"3.54.5"},"reference-count":15,"publisher":"IEEE","license":[{"start":{"date-parts":[[2026,5,18]],"date-time":"2026-05-18T00:00:00Z","timestamp":1779062400000},"content-version":"stm-asf","delay-in-days":0,"URL":"https:\/\/doi.org\/10.15223\/policy-029"},{"start":{"date-parts":[[2026,5,18]],"date-time":"2026-05-18T00:00:00Z","timestamp":1779062400000},"content-version":"stm-asf","delay-in-days":0,"URL":"https:\/\/doi.org\/10.15223\/policy-037"}],"funder":[{"DOI":"10.13039\/501100001321","name":"National Research Foundation","doi-asserted-by":"publisher","id":[{"id":"10.13039\/501100001321","id-type":"DOI","asserted-by":"publisher"}]}],"content-domain":{"domain":[],"crossmark-restriction":false},"short-container-title":[],"published-print":{"date-parts":[[2026,5,18]]},"DOI":"10.1109\/infocom59046.2026.11571434","type":"proceedings-article","created":{"date-parts":[[2026,6,29]],"date-time":"2026-06-29T19:38:15Z","timestamp":1782761895000},"page":"1-6","source":"Crossref","is-referenced-by-count":0,"title":["SLA-Aware Distributed LLM Inference Across Device-RAN-Cloud"],"prefix":"10.1109","author":[{"given":"Hariz","family":"Yet","sequence":"first","affiliation":[{"name":"Singapore University of Technology and Design (SUTD)"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Nguyen Thanh","family":"Tam","sequence":"additional","affiliation":[{"name":"Singapore University of Technology and Design (SUTD)"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Mao V.","family":"Ngo","sequence":"additional","affiliation":[{"name":"Singapore University of Technology and Design (SUTD)"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Lim Yi","family":"Shen","sequence":"additional","affiliation":[{"name":"Singapore University of Technology and Design (SUTD)"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Lin","family":"Wei","sequence":"additional","affiliation":[{"name":"Singapore University of Technology and Design (SUTD)"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Jihong","family":"Park","sequence":"additional","affiliation":[{"name":"Singapore University of Technology and Design (SUTD)"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Binbin","family":"Chen","sequence":"additional","affiliation":[{"name":"Singapore University of Technology and Design (SUTD)"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Tony Q. S.","family":"Quek","sequence":"additional","affiliation":[{"name":"Singapore University of Technology and Design (SUTD)"}],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"263","reference":[{"key":"ref1","first-page":"1","article-title":"AutoRAN: Automated and Zero-Touch Open RAN Systems","volume-title":"IEEE Transactions on Mobile Computing","author":"Maxenti","year":"2026"},{"key":"ref2","doi-asserted-by":"publisher","DOI":"10.1109\/jsac.2023.3334610"},{"key":"ref3","article-title":"Beyond Connectivity: An Open Architecture for AI-RAN Convergence in 6G","author":"Polese","year":"2025"},{"key":"ref4","article-title":"Ai-ran alliance vision and mission white paper","volume-title":"AIRAN Alliance, Tech. Rep.","year":"2024"},{"key":"ref5","doi-asserted-by":"publisher","DOI":"10.1109\/GLOBECOM59602.2025.11432469"},{"key":"ref6","doi-asserted-by":"publisher","DOI":"10.1109\/MCOM.001.2500018"},{"key":"ref7","doi-asserted-by":"publisher","DOI":"10.1145\/3718958.3750517"},{"key":"ref8","doi-asserted-by":"publisher","DOI":"10.1109\/INFOCOM52122.2024.10621380"},{"key":"ref9","doi-asserted-by":"publisher","DOI":"10.1109\/TC.2025.3604473"},{"key":"ref10","doi-asserted-by":"publisher","DOI":"10.52202\/068431-2198"},{"key":"ref11","article-title":"Gptq: Accurate post-training quantization for generative pre-trained transformers","author":"Frantar","year":"2023"},{"key":"ref12","article-title":"Awq: Activation-aware weight quantization for llm compression and acceleration","author":"Lin","year":"2024"},{"key":"ref13","article-title":"Nvidia multi-instance gpu (mig) user guide","year":"2023"},{"key":"ref14","article-title":"Qwen2.5-VL-3B-Instruct model card on hugging face","year":"2025","journal-title":"Hugging Face repository: huggingface.co\/Qwen\/Qwen2.5-VL-3B-Instruct"},{"key":"ref15","article-title":"Time-slicing gpus in kubernetes","volume-title":"nVIDIA GPU Operator Documentation (v24.3.0)","year":"2024"}],"event":{"name":"IEEE INFOCOM 2026 - IEEE Conference on Computer Communications","location":"Tokyo, Japan","start":{"date-parts":[[2026,5,18]]},"end":{"date-parts":[[2026,5,21]]}},"container-title":["IEEE INFOCOM 2026 - IEEE Conference on Computer Communications"],"original-title":[],"link":[{"URL":"http:\/\/xplorestaging.ieee.org\/ielx8\/11571071\/11571169\/11571434.pdf?arnumber=11571434","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,6,30]],"date-time":"2026-06-30T05:26:19Z","timestamp":1782797179000},"score":1,"resource":{"primary":{"URL":"https:\/\/ieeexplore.ieee.org\/document\/11571434\/"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,5,18]]},"references-count":15,"URL":"https:\/\/doi.org\/10.1109\/infocom59046.2026.11571434","relation":{},"subject":[],"published":{"date-parts":[[2026,5,18]]}}}