학술논문

HyperCLOVA X Technical Report
Document Type
Working Paper
Author
Yoo, Kang MinHan, JaegeunIn, SookyoJeon, HeewonJeong, JisuKang, JaewookKim, HyunwookKim, Kyung-MinKim, MunhyongKim, SungjuKwak, DonghyunKwak, HanockKwon, Se JungLee, BadoLee, DongsooLee, GichangLee, JoohoPark, BaeseongShin, SeongjinYu, JoonsangBaek, SeolkiByeon, SuminCho, EungsupChoe, DooseokHan, JeesungJin, YoungkyunJun, HyeinJung, JaeseungKim, ChanwoongKim, JinhongKim, JinukLee, DokyeongPark, DongwookSohn, Jeong MinHan, SujungHeo, JiaeHong, SungjuJeon, MinaJung, HyunhoonJung, JungeunJung, WangkyoKim, ChungjoonKim, HyeriKim, JonghyunKim, Min YoungLee, SoeunPark, JoonheeShin, JieunYang, SojinYoon, JungsoonLee, HwaranBae, SanghwanCha, JeehwanGylleus, KarlHam, DonghoonHong, MihakHong, YoungkiHong, YunkiJang, DahyunJeon, HyojunJeon, YujinJeong, YejiJi, MyunggeunJin, YegukJo, ChansongJoo, ShinyoungJung, SeunghwanKim, Adrian JungmyungKim, Byoung HoonKim, HyominKim, JungwhanKim, MinkyoungKim, MinseungKim, SungdongKim, YongheeKim, YoungjunKim, YoungkwanKo, DonghyeonLee, DughyunLee, Ha YoungLee, JaehongLee, JieunLee, JonghyunLee, JongjinLee, Min YoungLee, YehbinMin, TaehongMin, YuriMoon, KiyoonOh, HyangnamPark, JaesunPark, KyuyonPark, YounghunSeo, HanbaeSeo, SeunghyunSim, MihyunSon, GyubinYeo, MattYeom, Kyung HoonYoo, WonjoonYou, MyunginAhn, DoheonAhn, HominAhn, JooheeAhn, SeongminAn, ChanwooAn, HyeryunAn, JunhoAn, Sang-MinByun, BoramByun, EunbinCha, JonghoChang, MinjiChang, SeunggyuCho, HaesongCho, YoungdoChoi, DalnimChoi, DaseulChoi, HyoseokChoi, MinseongChoi, SanghoChoi, SeongjaeChoi, WooyongChun, SewhanGo, Dong YoungHam, ChiheonHan, DanbiHan, JaeminHong, MoonyoungHong, Sung BumHwang, Dong-HyunHwang, SeongchanIm, JinbaeJang, Hyuk JinJang, JaehyungJang, JaeniJang, SihyeonJang, SungwonJeon, JoonhaJeong, DaunJeong, JoonhyunJeong, KyeongseokJeong, MiniJin, SolJo, HanbyeolJo, HanjuJo, MinjungJung, ChaeyoonJung, HyungsikJung, JaeukJung, Ju HwanJung, KwangsunJung, SeungjaeKa, SoonwonKang, DonghanKang, SoyoungKil, TaehoKim, AreumKim, BeomyoungKim, ByeongwookKim, DaeheeKim, Dong-GyunKim, DonggookKim, DonghyunKim, EunaKim, EunchulKim, GeewookKim, Gyu RiKim, HanbyulKim, HeesuKim, IsaacKim, JeonghoonKim, JihyeKim, JoonghoonKim, MinjaeKim, MinsubKim, Pil HwanKim, SammyKim, SeokhunKim, SeonghyeonKim, SoojinKim, SoongKim, SoyoonKim, SunyoungKim, TaehoKim, WonhoKim, YoonsikKim, You JinKim, YuriKwon, BeomseokKwon, OhsungKwon, Yoo-HwanLee, AnnaLee, ByungwookLee, ChanghoLee, DaunLee, DongjaeLee, Ha-RamLee, HodongLee, HwiyeongLee, HyunmiLee, InjaeLee, JaeungLee, JeongsangLee, JisooLee, JongsooLee, JoongjaeLee, JuhanLee, Jung HyunLee, JunghoonLee, JunwooLee, Se YunLee, SujinLee, SungjaeLee, SungwooLee, WonjaeLee, Zoo HyunLim, Jong KunLim, KunLim, TaeminNa, NuriNam, JeongyeonNam, Kyeong-MinNoh, YeonseogOh, BiroOh, Jung-SikOh, SolgilOh, YeontaekPark, BoyounPark, CheonbokPark, DongjuPark, HyeonjinPark, Hyun TaePark, HyunjungPark, JihyePark, JooseokPark, JunghwanPark, JungsooPark, MiruPark, Sang HeePark, SeunghyunPark, SoyoungPark, TaerimPark, WonkyeongRyu, HyunjoonRyu, JeonghunRyu, NahyeonSeo, SoonshinSeo, Suk MinShim, YoonjeongShin, KyuyongShin, WonkwangSim, HyunSim, WoongseobSoh, HyejinSon, BokyongSon, HyunjunSon, SeulahSong, Chi-YunSong, ChiyoungSong, Ka YeonSong, MinchulSong, SeungminWang, JisungYeo, YonggooYi, Myeong YeonYim, Moon BinYoo, TaehwanYoo, YoungjoonYoon, SungminYoon, Young JinYu, HangyeolYu, Ui SeonZuo, XingdongBae, JeonginBae, JoungeunCho, HyunsooCho, SeonghyunCho, YongjinChoi, TaekyoonChoi, YeraChung, JiwanHan, ZhenghuiHeo, ByeonghoHong, EuisukHwang, TaebaekIm, SeonyeolJegal, SuminJeon, SuminJeong, YelimJeong, YonghyunJiang, CanJiang, JuyongJin, JihoJo, AraJo, YounghyunJung, HoyounJung, JuyoungKang, SeunghyeongKim, Dae HeeKim, GinamKim, HangyeolKim, HeeseungKim, HyojinKim, HyojunKim, Hyun-AhKim, JeehyeKim, Jin-HwaKim, JiseonKim, JonghakKim, Jung YoonKim, Rak YeongKim, SeongjinKim, SeoyoonKim, SewonKim, SooyoungKim, SukyoungKim, TaeyongKo, NaeunKoo, BonseungKwak, HeeyoungKwon, HaenaKwon, YoungjinLee, BoramLee, Bruce W.Lee, DagyeongLee, ErinLee, EuijinLee, Ha GyeongLee, HyojinLee, HyunjeongLee, JeeyoonLee, JeonghyunLee, JongheokLee, JoonhyungLee, JunhyukLee, MinguLee, NayeonLee, SangkyuLee, Se YoungLee, SeulgiLee, Seung JinLee, SuhyeonLee, YeonjaeLee, YesolLee, YoungbeomLee, YujinLi, ShaodongLiu, TianyuMoon, Seong-EunMoon, TaehongNihlenramstroem, Max-LasseOh, WonseokOh, YuriPark, HongbeenPark, HyekyungPark, JaehoPark, NohilPark, SangjinRyu, JiwonRyu, MiruRyu, SimoSeo, AhreumSeo, HeeSeo, KangdeokShin, JaminShin, SeungyounSin, HeetaeWang, JiangpingWang, LeiXiang, NingXiao, LongxiangXu, JingYi, SeonyeongYoo, HaanjuYoo, HaneulYoo, HwanheeYu, LiangYu, YoungjaeYuan, WeijieZeng, BoZhou, QianCho, KyunghyunHa, Jung-WooPark, JoonsukHwang, JihyunKwon, Hyoung JoKwon, SoonyongLee, JungyeonLee, SeunghoLim, SeonghyeonNoh, HyunkyungChoi, SeunghoLee, Sang-WooLim, Jung HwaSung, Nako
Source
Subject
Computer Science - Computation and Language
Computer Science - Artificial Intelligence
Language
Abstract
We introduce HyperCLOVA X, a family of large language models (LLMs) tailored to the Korean language and culture, along with competitive capabilities in English, math, and coding. HyperCLOVA X was trained on a balanced mix of Korean, English, and code data, followed by instruction-tuning with high-quality human-annotated datasets while abiding by strict safety guidelines reflecting our commitment to responsible AI. The model is evaluated across various benchmarks, including comprehensive reasoning, knowledge, commonsense, factuality, coding, math, chatting, instruction-following, and harmlessness, in both Korean and English. HyperCLOVA X exhibits strong reasoning capabilities in Korean backed by a deep understanding of the language and cultural nuances. Further analysis of the inherent bilingual nature and its extension to multilingualism highlights the model's cross-lingual proficiency and strong generalization ability to untargeted languages, including machine translation between several language pairs and cross-lingual inference tasks. We believe that HyperCLOVA X can provide helpful guidance for regions or countries in developing their sovereign LLMs.
Comment: 44 pages; updated authors list and fixed author names