Nduzi Asụsụ AI

Nucleus na Top-k Sampling

Nucleus (top-p) na top-k sampling bụ ụzọ ngbanwe nke na-agbakwunye enweghị usoro a na-achịkwa na ọgbọ ederede site na ịmachi akara nke enwere ike ịhọrọ.

2 nkeji na-agụEmelitere ikpeazụ

Nchịkọta

They matter because they make AI writing feel natural and varied instead of repetitive or robotic.

Ime miri emi

Ụdị asụsụ na-ewepụta nkesa ihe gbasara nke puru omume n'okwu ya dum n'usoro ọ bụla. Nlele anya ozugbo site na ya nwere ike nweta akara ngosi dị egwu, nke nwere ike ịdị ala; mgbe niile na-ewere top token (anyaukwu) na-amịpụta dull, ugboro ugboro loops. Top-k sampling na-edozi nke a site n'idebe naanị akara puru omume k (sị k=40), imezigharị, na nlele n'etiti ha. Nucleus sampling, nke Holtzman et al webatara. na 2019, kama na-edobe akara ngosi kacha nta nke ikike mkpokọta ha karịrị p (dịka ọmụmaatụ, 0.9) - 'nucleus'. Isi uru bụ na ntọala a na-ebelata mgbe ihe nlereanya nwere ntụkwasị obi ma gbasaa mgbe ọ na-ejighị n'aka, na-emegharị n'ike. A na-ejikọta ha abụọ na oke okpomọkụ nke na-amụba ma ọ bụ gbasaa nkesa tupu ịlele.

Nghọta nka nka

Ọdịiche dị oke mkpa bụ nke edobere na mbelata ngbanwe. Top-k na-edobe akara ngosi k, nke nwere ike ịdị ole na ole mgbe ọtụtụ nhọrọ nwere ezi uche, ma ọ bụ gụnyere junk mgbe naanị di na nwunye nwere ezi uche. Top-p na-edobe ọnụọgụ ọnụọgụ - naanị akara zuru oke iji kpuchie oke nke puru omume p - yabụ ọ na-akụda ọdụ ogologo a na-apụghị ịdabere na ya ma na-akwanyere etu nkesa ahụ si gbakọọ ma ọ bụ kwụ ọtọ. Okpomọkụ (nke na-abụkarị 0.7-1.0) na-eweghachite logits tupu usoro ọ bụla: ụkpụrụ dị ala na-etinye uche na puru omume, ụkpụrụ dị elu na-agbasa ya.

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 Nucleus na Top-k Sampling

Nchọpụta nke dabere na nlele ugbu a bụ ihe ndabara maka nkata na ngwaọrụ okike, nyocha na-aga n'ihu na-emezigharị ya: ụzọ dị ka nlele ụdị, min-p, na eta/epsilon sampling ebumnuche iji mezie ọdụ ọdụ ahụ nke ọma karịa p ma ọ bụ k. Na-atụ anya ngbanwe ngbanwe ka ọ bụrụ ịmatakwu ihe ndị gbara ya gburugburu na ọbụna mụta, na-akpachi anya na-akpaghị aka maka azịza eziokwu yana ịtọghe maka ịtụgharị uche. Ka ụdịdị na-akawanye mma, njikwa nleba anya nke ọma ka dị mkpa maka ịhazi ntụkwasị obi, ụdị dị iche iche, na ibelata nhụsianya.

Mmejuputa n'ezie n'ụwa

Chatbots na-eji top-p gburugburu 0.9 iji nọgide na-aza azịza dị iche iche ma na-adakọ ọnụ na mkparịta ụka.

Ndị na-enyere aka ide ihe okike na-ebuli okpomọkụ na p iji chepụta echiche akụkọ dị iche iche

Ngwa-ọgbọ koodu na-eweda ọnọdụ okpomọkụ yana k maka mkpebi siri ike karị, snippets ziri ezi

Ndị ọrụ API na-emegharị top_p na top_k paramita iji jikwaa ka mpụta ihe nlereanya si dị egwu

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 Nucleus and Top-k Sampling 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

Nlele a na-ahụkarị

Ajụjụ a na-ajụkarị

What is Nucleus and Top-k Sampling?

Nucleus (top-p) na top-k sampling bụ ụzọ ngbanwe nke na-agbakwunye enweghị usoro a na-achịkwa na ọgbọ ederede site na ịmachi akara nke enwere ike ịhọrọ. Ha dị mkpa n'ihi na ha na-eme ka ide AI nwee mmetụta nke okike ma dị iche iche kama ịmegharị ma ọ bụ roboti.

Kedu ka nlele top-k si egbochi nhọrọ token?

Top-k na-ewepụ nkesa na nọmba edobere k nke akara ndị nwere ike ịmegharị, na-emegharị, na ihe nlele sitere na ndị ahụ.

Kedu ihe na-akọwa 'nucleus' na nucleus (top-p) sample?

Top-p na-edobe otu kacha nta nke akara ngosi kacha elu nke ihe gbasara ike ha ruru n'ọnụ ụzọ p, wee lelee n'etiti ha.

Kedu uru bụ isi nke top-p karịa top-k?

Onye ndoro-ndoro ochichi Top-p na-agbada mgbe ihe nlereanya nwere ntụkwasị obi wee tolite mgbe ejighị n'aka, n'adịghị ka ọnụọgụ nke top-k.

Kedu ihe oke okpomọkụ na-eme tupu ịlele?

Obere okpomọkụ na-etinye uche na ihe gbasara nke puru omume na tokens dị elu (karịa deterministic); okpomọkụ dị elu na-eme ka ọ daa (karịa iche iche).

Kedu ihe kpatara ngbanwe nke anyaukwu dị ọcha (mgbe niile bụ akara ngosi kachasị elu) anaghị achọkarị ederede mepere emepe?

Ndozi anya anya na-arapara n'ahụ ugboro ugboro na enweghị ụdị okike dị iche iche nke na-achịkwa nlele.