Guhuza amabwiriza
Guhuza amabwiriza ni intambwe yo guhugura ihindura inyandiko-iteganya kuba icyitegererezo gikurikiza amabwiriza nka 'incamake ibi' cyangwa 'andika igisubizo kinyabupfura.' Nibyo bituma icyitegererezo fatizo cyumva gifasha kandi kiyobora.
Kwibira cyane
Ururimi rwibanze rwatojwe gusa guhanura ibimenyetso bizakurikira kumyandiko y'urubuga, niba rero wanditse ikibazo birashobora gukomeza nibibazo byinshi aho gusubiza. Guhindura amabwiriza bikosora ibi. Nuburyo bwo kugenzurwa neza-gutunganya neza: icyitegererezo cyatojwe kubice byinshi bya (amabwiriza, igisubizo cyiza) gikubiyemo imirimo ibihumbi - ubusemuzi, incamake, ibyiciro, Q&A, code, nibindi byinshi. Kubona amabwiriza amwe-hanyuma-afasha-gusubiza inshuro nyinshi, icyitegererezo cyiga imyitwarire rusange yo 'gukora ibyo umukoresha asabye,' kandi ibi bihurira kumabwiriza atigeze abona mumahugurwa. Ubu buryo bwashyizweho ahagana mu 2021 nakazi nka FLAN, T0, n’Amabwiriza Kamere, kandi byari shingiro rya __AIU_PROTECTED_10 __ 'InstructGPT, ihuza neza GPT-3 kumurongo wateganijwe. Nishingiro abafasha benshi baganira bubakiyeho.
Ubushishozi
Muburyo bwa tekinike, guhuza amabwiriza ni bisanzwe bigenzurwa no kwiga: gabanya itandukaniro riri hagati yikimenyetso cyerekana icyitegererezo hamwe nigisubizo cyerekanwe, hamwe na gradients ivugurura uburemere. Iratandukanye na RLHF (gushimangira kwigira kubitekerezo byabantu), biza nyuma kandi bigahindura ibyifuzo byabantu ukoresheje icyitegererezo. Ibisanzwe bisanzwe byateguwe: kwitegura, hanyuma amabwiriza-tune (SFT) kugirango wigishe gukurikiza imirimo, hanyuma ubishaka RLHF kunonosora amajwi, ubufasha, numutekano. Itandukaniro ryamakuru rifite akamaro kuruta ubwinshi - gukwirakwiza imirimo yagutse itera rusange.
Ingaruka z'Ingamba
Umuvuduko n'igipimo
Ururimi rwakazi rushobora kugenda byihuse nta gutamba guhuzagurika.
Kugera no kugera
Yagura uburyo bwindimi nuburyo bwo gutumanaho.
Ibyemezo bisobanutse
Amakipe arashobora kumara umwanya munini murubanza mugihe automatike ikora gusubiramo.
Ejo hazaza h'inyigisho
Umwanya urimo uva mubitabo binini byandikishijwe intoki werekeza kumurongo wohejuru, igice cyogukora amakuru - rimwe na rimwe ibihumbi bike gusa byatoranijwe neza - nyuma yubushakashatsi bwerekana ko ubwiza bwamakuru bushobora gutsinda ubwinshi. Witege byinshi kumurongo wihariye woguhuza amabwiriza (ubuvuzi, amategeko, coding), indimi nyinshi nindimi nyinshi, hamwe numuyoboro wikora utanga kandi ushungura amakuru yubuyobozi. Guhindura amabwiriza bizagumaho ikiraro cyingenzi hagati yicyitegererezo cyateganijwe mbere yumufasha wifashishwa, bigenda bihuzwa no guhitamo neza guhuza.
Gushyira mu bikorwa Isi
Guhindura ishingiro rya GPT-muburyo bwumufasha wo kuganira asubiza ibibazo aho kubisubiramo
FLAN-T5, itunganijwe neza mubikorwa byinshi kuburyo ishobora gukurikiza amabwiriza itigeze itozwa neza
InstructGPT, aho GPT-3 yari amabwiriza-yashizwe kumurongo wateguwe kugirango utange ibisubizo byingirakamaro cyane
Kubaka umufasha wikigo imbere muguhuza neza amabwiriza-ibisubizo byombi byanditswe ninkunga nitsinda ryemewe
Ingaruka & Kurinda
Ibintu bifatika bishobora kwinjiza bucece raporo, gushyigikira imigendekere, cyangwa ibisubizo byubushakashatsi.
Kwihuta byihuse birashobora gukora ibisubizo bidahuye mubisabwa bisa.
Ibyanditswe byumvikana birashobora kugaragara niba kugenzura kugenzura ari ntege.
Igishushanyo mbonera
Sobanura imiterere isohoka, amajwi, hamwe nubuziranenge mbere yo gutangira.
Ibisubizo byibanze hamwe nisoko yizewe igihe cyose ukuri kwingirakamaro.
Komeza kugenzura abantu kugenzura ibisubizo byinshi.
Kurikirana uburyo bwo kunanirwa no kongera imyitozo cyangwa akazi gahoraho.
Komeza Ubushakashatsi
Free newsletter
Get the daily AI briefing
Three verified AI stories every weekday morning, written in plain English. Free forever, no ads.
One email each weekday. Unsubscribe in one click. We never sell or share your address.
Test yourself
Take the Instruction Tuning quiz
Instant feedback on every answer, and a shareable certificate with a verifiable ID once you pass a course.
Support free AI education. AI Understanding is a 501(c)(3) nonprofit — no ads, no paywall, ever. Make a donation
Ubuyobozi bukurikira
Guhuza imbanzirizamushinga
Ibibazo bikunze kubazwa
What is Instruction Tuning?
Guhuza amabwiriza ni intambwe yo guhugura ihindura inyandiko-iteganya kuba icyitegererezo gikurikiza amabwiriza nka 'incamake ibi' cyangwa 'andika igisubizo kinyabupfura.' Nibyo bituma icyitegererezo fatizo cyumva gifasha kandi kiyobora.
Guhuza amabwiriza bitandukaniye he na RLHF?
Guhuza amabwiriza bigenzurwa no kwiga kubisubizo bigamije. RLHF iza nyuma hanyuma igahindura icyitegererezo kirwanya ibyo abantu bize.
Nuwuhe mutungo winyigisho-uhuza amakuru menshi atwara ubushobozi bwikitegererezo gukurikiza amabwiriza mashya, atagaragara?
Gupfukirana imirimo itandukanye yigisha imyitwarire rusange yamabwiriza akurikira, ashyira mubikorwa amabwiriza atigeze aboneka mumahugurwa.
Nubuhe buryo busanzwe bwo gutoza ibyiciro byumufasha wa kijyambere?
Abanyamideli babanje gutegurwa kumyandiko mbisi, hanyuma amabwiriza-akurikiranwa kugirango akurikire imirimo, kandi akenshi atunganywa na RLHF kugirango ijwi n'umutekano.