[{"data":1,"prerenderedAt":282},["ShallowReactive",2],{"content-doc-\u002Fzh\u002Fnews\u002F20261010":3},{"_path":4,"_dir":5,"_draft":6,"_partial":6,"_locale":7,"title":8,"description":7,"date":9,"author":10,"tags":11,"body":14,"_type":276,"_id":277,"_source":278,"_file":279,"_stem":280,"_extension":281},"\u002Fzh\u002Fnews\u002F20261010","news",false,"","低精度计算亮相CNCC：数据格式、算法与系统协同的新前沿","2026\u002F10\u002F10","全球计算联盟GCC",[12,13],"CNCC2026","论坛",{"type":15,"children":16,"toc":272},"root",[17,29,33,39,44,51,250,260],{"type":18,"tag":19,"props":20,"children":22},"element","pre",{"code":21},"大模型时代，如何用更少的比特获得更高的计算效率？GCC-HiFloat社区将于10月22日在CNCC 2026大会举办分论坛：低精度计算论坛，聚焦低精度数据格式、模型训练与推理、量化算法、软硬件协同及系统优化等前沿研究，探讨从数值表示创新到高效AI计算系统构建的关键问题与最新进展。\n",[23],{"type":18,"tag":24,"props":25,"children":26},"code",{"__ignoreMap":7},[27],{"type":28,"value":21},"text",{"type":18,"tag":30,"props":31,"children":32},"br",{},[],{"type":18,"tag":34,"props":35,"children":36},"p",{},[37],{"type":28,"value":38},"但低精度不只是减少几个bit，位宽不断降低后，模型精度、数值动态范围、量化误差、训练稳定性以及硬件执行效率之间的矛盾更加突出。如何设计更适合AI数据分布的新型低精度数据格式，如何实现训练、推理和量化算法协同，如何通过编译、算子与体系结构设计将理论上的位宽优势真正转化为端到端性能收益，仍存在大量值得深入研究的问题。",{"type":18,"tag":34,"props":40,"children":41},{},[42],{"type":28,"value":43},"本论坛将面向低精度计算的学术前沿，围绕低精度数值表示与数据格式、低比特模型训练与推理、量化算法、软硬件协同设计、高效算子与体系结构、低精度计算系统与数据传输等方向展开交流，邀请国内相关领域专家学者分享最新研究成果，并共同探讨低精度计算未来值得关注的基础问题、关键技术与研究趋势。",{"type":18,"tag":45,"props":46,"children":48},"h2",{"id":47},"论坛议程",[49],{"type":28,"value":50},"论坛议程：",{"type":18,"tag":52,"props":53,"children":54},"table",{},[55,84],{"type":18,"tag":56,"props":57,"children":58},"thead",{},[59],{"type":18,"tag":60,"props":61,"children":62},"tr",{},[63,69,74,79],{"type":18,"tag":64,"props":65,"children":66},"th",{},[67],{"type":28,"value":68},"序号",{"type":18,"tag":64,"props":70,"children":71},{},[72],{"type":28,"value":73},"主题",{"type":18,"tag":64,"props":75,"children":76},{},[77],{"type":28,"value":78},"主讲嘉宾",{"type":18,"tag":64,"props":80,"children":81},{},[82],{"type":28,"value":83},"单位",{"type":18,"tag":85,"props":86,"children":87},"tbody",{},[88,112,135,158,181,204,227],{"type":18,"tag":60,"props":89,"children":90},{},[91,97,102,107],{"type":18,"tag":92,"props":93,"children":94},"td",{},[95],{"type":28,"value":96},"1",{"type":18,"tag":92,"props":98,"children":99},{},[100],{"type":28,"value":101},"低精算子开发与性能优化",{"type":18,"tag":92,"props":103,"children":104},{},[105],{"type":28,"value":106},"陆璐",{"type":18,"tag":92,"props":108,"children":109},{},[110],{"type":28,"value":111},"华南理工大学",{"type":18,"tag":60,"props":113,"children":114},{},[115,120,125,130],{"type":18,"tag":92,"props":116,"children":117},{},[118],{"type":28,"value":119},"2",{"type":18,"tag":92,"props":121,"children":122},{},[123],{"type":28,"value":124},"大模型数值精度：误解、原理与瓶颈",{"type":18,"tag":92,"props":126,"children":127},{},[128],{"type":28,"value":129},"尚笠",{"type":18,"tag":92,"props":131,"children":132},{},[133],{"type":28,"value":134},"复旦大学",{"type":18,"tag":60,"props":136,"children":137},{},[138,143,148,153],{"type":18,"tag":92,"props":139,"children":140},{},[141],{"type":28,"value":142},"3",{"type":18,"tag":92,"props":144,"children":145},{},[146],{"type":28,"value":147},"人工智能计算系统设计及优化",{"type":18,"tag":92,"props":149,"children":150},{},[151],{"type":28,"value":152},"王培松",{"type":18,"tag":92,"props":154,"children":155},{},[156],{"type":28,"value":157},"中国科学院自动化所",{"type":18,"tag":60,"props":159,"children":160},{},[161,166,171,176],{"type":18,"tag":92,"props":162,"children":163},{},[164],{"type":28,"value":165},"4",{"type":18,"tag":92,"props":167,"children":168},{},[169],{"type":28,"value":170},"面向AI计算的低精度数据格式的软硬件协同优化",{"type":18,"tag":92,"props":172,"children":173},{},[174],{"type":28,"value":175},"鲁金铭",{"type":18,"tag":92,"props":177,"children":178},{},[179],{"type":28,"value":180},"南京大学",{"type":18,"tag":60,"props":182,"children":183},{},[184,189,194,199],{"type":18,"tag":92,"props":185,"children":186},{},[187],{"type":28,"value":188},"5",{"type":18,"tag":92,"props":190,"children":191},{},[192],{"type":28,"value":193},"极低比特大模型推理：从数值表示到高效执行",{"type":18,"tag":92,"props":195,"children":196},{},[197],{"type":28,"value":198},"冷静文",{"type":18,"tag":92,"props":200,"children":201},{},[202],{"type":28,"value":203},"上海交通大学",{"type":18,"tag":60,"props":205,"children":206},{},[207,212,217,222],{"type":18,"tag":92,"props":208,"children":209},{},[210],{"type":28,"value":211},"6",{"type":18,"tag":92,"props":213,"children":214},{},[215],{"type":28,"value":216},"迈向极致低精度的基础模型：从数值、张量到时序",{"type":18,"tag":92,"props":218,"children":219},{},[220],{"type":28,"value":221},"秦浩桐",{"type":18,"tag":92,"props":223,"children":224},{},[225],{"type":28,"value":226},"香港中文大学",{"type":18,"tag":60,"props":228,"children":229},{},[230,235,240,245],{"type":18,"tag":92,"props":231,"children":232},{},[233],{"type":28,"value":234},"7",{"type":18,"tag":92,"props":236,"children":237},{},[238],{"type":28,"value":239},"Hifloat格式介绍及训推实践",{"type":18,"tag":92,"props":241,"children":242},{},[243],{"type":28,"value":244},"李思坦",{"type":18,"tag":92,"props":246,"children":247},{},[248],{"type":28,"value":249},"华为技术有限公司",{"type":18,"tag":34,"props":251,"children":252},{},[253,255,258],{"type":28,"value":254},"会议时间：10月22日13:30~17:30",{"type":18,"tag":30,"props":256,"children":257},{},[],{"type":28,"value":259},"\n会议地点：成都天府国际会议中心-107A三星堆A会议室",{"type":18,"tag":34,"props":261,"children":262},{},[263,265],{"type":28,"value":264},"参会联系：",{"type":18,"tag":266,"props":267,"children":269},"a",{"href":268},"mailto:guizelin@gccorg.com",[270],{"type":28,"value":271},"guizelin@gccorg.com",{"title":7,"searchDepth":273,"depth":273,"links":274},2,[275],{"id":47,"depth":273,"text":50},"markdown","app:zh:news:20261010.md","app","zh\u002Fnews\u002F20261010.md","zh\u002Fnews\u002F20261010","md",1795070546054]