Nduzi Asụsụ AI

Nlereanya na nhazi ọkwa kacha mma

Nlereanya kacha mma na-ewepụta azịza dị iche iche site na ụdị wee họrọ nke kacha mma site na iji usoro akara dị iche.

2 nkeji na-agụEmelitere ikpeazụ

Nchịkọta

It is one of the simplest, most reliable ways to trade extra compute at inference time for higher answer quality.

Ime miri emi

Ụdị asụsụ nwere nlele anya na-ewepụta mpụta dị iche iche oge ọ bụla ị na-eme ya. Best-of-N na-erigbu nke a: ị na-eseta nzaghachi onye ndoro-ndoro ochichi N, wee degharịa ha ma weghachi nke kachasị elu. Onye nnochite anya nwere ike ịbụ ihe nlere nkwụghachi amụtara (nke a na-ahụkarị na mmụta nkwado sitere na nzaghachi mmadụ), onye nyocha na-enyocha izi ezi, ma ọ bụ nkwekọrịta dị mfe dị ka nkwekọrịta azịza site na ịtụ vootu. N'ihi na ihe nlereanya ahụ chọrọ naanị otu ezigbo mbọ n'ime ọtụtụ ndị, ịdịmma na-ebili nke ọma ka N na-etolite, ọkachasị na ntụgharị uche na koodu ọrụ ebe ụzọ ziri ezi dị mana ọ bụghị mgbe niile ka ọ bụrụ ihe nlele mbụ. Ọnụ ego ahụ bụ ahịrị na N, ma nwetakwa n'ikpeazụ dị larịị ma ọ bụ ọbụna tụgharịa ma ọ bụrụ na onye akara ezughị okè, ọnọdụ ọdịda nke a na-akpọ hacking ụgwọ ọrụ ma ọ bụ ụgwọ ọrụ karịa.

Nghọta nka nka

Ogo nke kacha mma-nke-N dabere kpamkpam na onye na-agba akara. Na a zuru okè verifier, ziri ezi na-abịaru nso ohere na ọ dịkarịa ala otu n'ime N samples ziri ezi, nke na-ebili ngwa ngwa na N. Na a ụzụ ụgwọ ọrụ nlereanya, nhọrọ nwere ike ịghọgbu: pushing N nnọọ elu amplifies mmepụta na akara elu ma n'ezie na-ezighị ezi, ebe ọ bụ na ị na- optimizing megide scorer si kpuru ntụpọ. Nke a bụ ya mere ụdị nkwụghachi ụgwọ akwụ ụgwọ siri ike dị mkpa maka usoro iji nọgide na-akwụ ụgwọ.

Mmetụta atụmatụ

Ọsọ na ọnụ ọgụgụ

Usoro ọrụ asụsụ nwere ike ịga ngwa ngwa n'achụghị nkwụsi ike.

Nweta na iru

Ọ na-agbasawanye ohere n'ofe asụsụ na ụdị nzikọrịta ozi.

Mkpebi doro anya

Ndị otu nwere ike itinyekwu oge na ikpe ebe akpaaka na-ejikwa nkwughachi.

Ọdịnihu nke nlele kacha mma na ntinye ọkwa

Best-of-N na-aghọ ihe bụ isi ihe na-ewu ụlọ nke oge ntule, n'akụkụ agbụ-echiche na ịchọ osisi. Na-atụ anya ụdị dị iche iche nwere ọgụgụ isi: ịtụ vootu dị arọ, usoro nkwụghachi ụgwọ ụdị nke na-egosi usoro ntụgharị uche ọ bụla, yana N na-akwụsị ịsepụta mgbe ntụkwasị obi dị elu. Ka verifiers na-akawanye mma, ọkachasị maka koodu na mgbakọ na mwepụ ebe enwere ike ịlele izizi, ịtụgharị ọtụtụ sample ga-abụ ụzọ ọkọlọtọ iji gbanwee mgbakọ mapụtara ka ọ bụrụ ntụkwasị obi na-azụghachighị ụdị ntọala ahụ.

Mmejuputa n'ezie n'ụwa

Ịtụle ihe ngwọta 64 maka nsogbu mgbakọ na mwepụ na ịhọpụta azịza nke ọtụtụ ihe atụ kwenyere na ya (ịkwụsi ike onwe onye / ọtụtụ ntuli aka).

Na-emepụta ọtụtụ koodu mmecha na idobe nke na-agafe ule otu kachasị dị ka ihe nnwapụta akpaka.

Inweta nzaghachi dị iche iche na pipeline RLHF yana ịhọrọ nzaghachi akara ngosi kacha elu-ụgwọ ọrụ iji jeere ndị ọrụ ozi.

Na-ewepụta ọtụtụ nchịkọta akwụkwọ na iji ụkpụrụ dị mma tinyegharịa ha ka ha weghachi nke kacha kwesị ntụkwasị obi, nke dị nkenke.

Ihe ize ndụ & okporo ụzọ nche

Eziokwu ndị e chepụtara echepụta nwere ike jiri nwayọ tinye akụkọ, nkwado nkwado, ma ọ bụ nsonaazụ nyocha.

Mmetụta ngwa ngwa nwere ike ịmepụta nsonaazụ na-ekwekọghị ekwekọ n'ofe arịrịọ ndị yiri ya.

Enwere ike ikpughe data ederede nwere mmetụta ma ọ bụrụ na njikwa ohere adịghị ike.

Map mmejuputa

1

Kọwaa usoro mmepụta, ụda, na ụkpụrụ ịdịmma tupu ibugharị.

2

Weghachite nzaghachi site na isi mmalite ntụkwasị obi mgbe ọ bụla izi ezi dị mkpa.

3

Debe ebe nleba anya mmadụ maka mpụta dị elu.

4

Sochie ụkpụrụ ọdịda ma na-azụghachi mkpali ma ọ bụ usoro ọrụ mgbe niile.

Nọgide na-eme nchọpụta

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 Best-of-N Sampling and Reranking quiz

Instant feedback on every answer, and a shareable certificate with a verifiable ID once you pass a course.

Malite ajụjụ

Support free AI education. AI Understanding is a 501(c)(3) nonprofit — no ads, no paywall, ever. Make a donation

Ntuziaka na-esote

Okpomọkụ na Sampling

Ajụjụ a na-ajụkarị

What is Best-of-N Sampling and Reranking?

Nlereanya kacha mma na-ewepụta azịza dị iche iche site na ụdị wee họrọ nke kacha mma site na iji usoro akara dị iche. Ọ bụ otu n'ime ụzọ kachasị mfe, nke a pụrụ ịdabere na ya iji zụọ ahịa mgbakwunye n'oge ntinye maka ogo azịza dị elu.

Gịnị bụ isi echiche nke kacha-nke-N sample?

Best-of-N na-adọta nzaghachi ndị ndoro-ndoro anya wee degharịa ha, na-eweghachi nke kacha nwee akara.

Kedu ụdị ọrụ kacha mma-nke-N na-akacha enyere aka?

Mgbe enwere azịza ziri ezi nke nwere ike ịnwapụta nke ihe nlereanya ahụ na-ahụ naanị mgbe ụfọdụ, ịchepụta ọtụtụ ndị na-aga ime na-ebuli ohere nke onye ahụ ziri ezi.

Kedu ihe na-ekpebi ma kacha mma-nke- na-eme ka nsonaazụ ya dịkwuo mma?

Nhọrọ dị naanị mma dị ka onye ntọgharị; onye na-agba akara na-adịghị ike ma ọ bụ nke enweghị mmasị nwere ike ịhọrọ azịza na-ezighi ezi.

Kedu ụdị ọdịda nwere ike ịpụta mgbe N na-eji ụdị ụgwọ ọrụ ezughị okè kwalie?

Ịmepụta ihe siri ike megide onye na-agba akara nwere ntụpọ na-abawanye mmepụta nke na-erigbu ntụpọ ya, nke mere na izi ezi nwere ike gbadaa ma ọ bụ daa.

Kedu atụmatụ ngbanwe dị mfe nke a makwaara dị ka nkwenye onwe onye?

Nkwenye onwe onye na-atụnye ọtụtụ agbụ echiche wee họrọ azịza ikpeazụ nke ọtụtụ n'agbụ kwekọrịtara na ya.