{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,3,3]],"date-time":"2026-03-03T23:37:10Z","timestamp":1772581030893,"version":"3.50.1"},"reference-count":23,"publisher":"IEEE","license":[{"start":{"date-parts":[[2024,9,23]],"date-time":"2024-09-23T00:00:00Z","timestamp":1727049600000},"content-version":"stm-asf","delay-in-days":0,"URL":"https:\/\/doi.org\/10.15223\/policy-029"},{"start":{"date-parts":[[2024,9,23]],"date-time":"2024-09-23T00:00:00Z","timestamp":1727049600000},"content-version":"stm-asf","delay-in-days":0,"URL":"https:\/\/doi.org\/10.15223\/policy-037"}],"funder":[{"DOI":"10.13039\/501100001810","name":"National Natural Science Foundation of China","doi-asserted-by":"publisher","award":["62325405,62104128,U21B2031,62204164"],"award-info":[{"award-number":["62325405,62104128,U21B2031,62204164"]}],"id":[{"id":"10.13039\/501100001810","id-type":"DOI","asserted-by":"publisher"}]}],"content-domain":{"domain":[],"crossmark-restriction":false},"short-container-title":[],"published-print":{"date-parts":[[2024,9,23]]},"DOI":"10.1109\/hpec62836.2024.10938498","type":"proceedings-article","created":{"date-parts":[[2025,4,3]],"date-time":"2025-04-03T19:07:19Z","timestamp":1743707239000},"page":"1-7","source":"Crossref","is-referenced-by-count":2,"title":["GLITCHES: GPU-FPGA LLM Inference Through a Collaborative Heterogeneous System"],"prefix":"10.1109","author":[{"given":"Fan","family":"Yang","sequence":"first","affiliation":[{"name":"Tsinghua University,BNRist,Dept. of EE"}],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"Xinhao","family":"Yang","sequence":"additional","affiliation":[{"name":"Tsinghua University,BNRist,Dept. of EE"}],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"Hongyi","family":"Wang","sequence":"additional","affiliation":[{"name":"Tsinghua University,BNRist,Dept. of EE"}],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"Zehao","family":"Wang","sequence":"additional","affiliation":[{"name":"Tsinghua University,BNRist,Dept. of EE"}],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"Zhenhua","family":"Zhu","sequence":"additional","affiliation":[{"name":"Tsinghua University,BNRist,Dept. of EE"}],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"Shulin","family":"Zeng","sequence":"additional","affiliation":[{"name":"Tsinghua University,BNRist,Dept. of EE"}],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"Yu","family":"Wang","sequence":"additional","affiliation":[{"name":"Tsinghua University,BNRist,Dept. of EE"}],"role":[{"role":"author","vocabulary":"crossref"}]}],"member":"263","reference":[{"key":"ref1","article-title":"Emergent abilities of large language models","author":"Wei","year":"2022","journal-title":"arXiv preprint"},{"key":"ref2","doi-asserted-by":"publisher","DOI":"10.1038\/s41591-023-02448-8"},{"key":"ref3","doi-asserted-by":"publisher","DOI":"10.1145\/3649506"},{"key":"ref4","article-title":"A complete survey on llm-based ai chatbots","author":"Dam","year":"2024","journal-title":"arXiv preprint"},{"key":"ref5","article-title":"Llm-empowered chatbots for psychiatrist and patient simulation: application and evaluation","author":"Chen","year":"2023","journal-title":"arXiv preprint"},{"key":"ref6","article-title":"Evaluating large language models trained on code","author":"Chen","year":"2021","journal-title":"arXiv preprint"},{"key":"ref7","doi-asserted-by":"publisher","DOI":"10.1145\/3597503.3639187"},{"key":"ref8","first-page":"12 098","article-title":"Longcoder: A long-range pre-trained language model for code completion","volume-title":"International Conference on Machine Learning. PMLR","author":"Guo","year":"2023"},{"key":"ref9","article-title":"A comprehensive overview of large language models","author":"Naveed","year":"2023","journal-title":"arXiv preprint"},{"key":"ref10","doi-asserted-by":"publisher","DOI":"10.1145\/3600006.3613165"},{"key":"ref11","doi-asserted-by":"publisher","DOI":"10.1109\/SC41404.2022.00051"},{"key":"ref12","author":"Vaidya","year":"2023","journal-title":"Optimizing inference on large language models with nvidia tensorrt-llm, now publicly available"},{"key":"ref13","doi-asserted-by":"publisher","DOI":"10.1109\/MICRO56248.2022.00051"},{"key":"ref14","doi-asserted-by":"publisher","DOI":"10.1145\/3626202.3637562"},{"key":"ref15","doi-asserted-by":"publisher","DOI":"10.48550\/ARXIV.1706.03762"},{"key":"ref16","doi-asserted-by":"publisher","DOI":"10.1145\/3656177"},{"key":"ref17","article-title":"Llama: Open and efficient foundation language models","author":"Touvron","year":"2023","journal-title":"arXiv preprint"},{"key":"ref18","first-page":"16344","article-title":"Flashattention: Fast and memory-efficient exact attention with io-awareness","volume":"35","author":"Dao","year":"2022","journal-title":"Advances in Neural Information Processing Systems"},{"key":"ref19","author":"Dao","year":"2023","journal-title":"Flash-decoding for long-context inference"},{"key":"ref20","first-page":"31 094","article-title":"Flexgen: High-throughput generative inference of large language models with a single gpu","volume-title":"International Conference on Machine Learning. PMLR","author":"Sheng","year":"2023"},{"key":"ref21","article-title":"Flashdecoding++: Faster large language model inference on gpus","author":"Hong","year":"2023","journal-title":"arXiv preprint"},{"key":"ref22","article-title":"The feasibility of implementing large-scale transformers on multi-fpga platforms","author":"Gao","year":"2024","journal-title":"arXiv preprint"},{"key":"ref23","doi-asserted-by":"publisher","DOI":"10.1109\/FCCM48280.2020.00024"}],"event":{"name":"2024 IEEE High Performance Extreme Computing Conference (HPEC)","location":"Wakefield, MA, USA","start":{"date-parts":[[2024,9,23]]},"end":{"date-parts":[[2024,9,27]]}},"container-title":["2024 IEEE High Performance Extreme Computing Conference (HPEC)"],"original-title":[],"link":[{"URL":"http:\/\/xplorestaging.ieee.org\/ielx8\/10938401\/10938415\/10938498.pdf?arnumber=10938498","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2025,4,4]],"date-time":"2025-04-04T11:15:13Z","timestamp":1743765313000},"score":1,"resource":{"primary":{"URL":"https:\/\/ieeexplore.ieee.org\/document\/10938498\/"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2024,9,23]]},"references-count":23,"URL":"https:\/\/doi.org\/10.1109\/hpec62836.2024.10938498","relation":{},"subject":[],"published":{"date-parts":[[2024,9,23]]}}}