Nlele a na-ahụkarị
Nlereanya a na-ahụkarị bụ usoro ọgbọ nke ederede na-ewere okwu na-esote site na tokens nke ọdịnaya ọdịnaya ha na-anọdụ ala na ihe ịtụnanya a tụrụ anya ka ihe nlereanya ahụ, kama ijide ndị kacha nwee ike mgbe niile.
Nchịkọta
It aims for output that feels natural and human-like by matching how real language balances predictability and novelty.
Ime miri emi
Mgbe ụdị asụsụ na-ebu amụma akara na-esote, ọ na-ewepụta nkesa puru omume karịa puku kwuru puku nhọrọ. Ụzọ anyaukwu na top-k na-akwado akara ngosi nwere ike dị elu, nke nwere ike ime ka ederede na-emegharị ugboro ugboro na nke na-adịghị mma. Nlereanya a na-ahụkarị, nke Meister na ndị ọrụ ibe webatara na 2022, na-ewe akụkụ dị iche gbanyere mkpọrọgwụ na usoro ozi. Ihe nlereanya ahụ na-agbakọ ọdịnaya ozi a na-atụ anya ya (entropy nke nkesa). A na-enweta akara ngosi site n'ókè ihe ịtụnanya nke ha si dị anya site na atụmanya ahụ. Nlereanya a na-ahụkarị na-edobe otu akara ngosi nke ihe ịtụnanya ya kacha nso na nkezi ruo mgbe ohere ha jikọtara ọnụ rute n'ọnụ ọnụ, wee nweta nlele sitere na ntọala ahụ. Ihe si na ya pụta bụ ederede nke na-awụ akpata oyi n'ahụ, ma ọ bụ nke a na-atụghị anya ya, na-egosipụta ka ụmụ mmadụ na-ekwurịta okwu na-adịkarị nso n'ogo ozi na-aga n'ihu.
Nghọta nka nka
Maka onye ndoro-ndoro ochichi nke ọ bụla, ihe nlereanya ahụ na-agbakọ ihe ịtụnanya, ihe puru omume na-adịghị mma. Ọ na-agbakọkwa entropy ọnọdụ, nkezi ihe puru omume-atụ n'anya karịa akara niile. Ụdị nlere anya n'ọkwa akara site na nnukwu ọdịiche dị n'etiti ihe ịtụnanya ha na nke ahụ entropy, wee jiri anyaukwu na-agbakwụnye akara kacha nso ruo mgbe ihe gbasara puru omume ha na-adaba na parameter tau (na-adịkarị gburugburu 0.9 ruo 0.95). Nlele na-eme naanị n'ime usoro a na-ahụkarị na mpaghara, na-egbochi ma ndị na-apụ apụ na nhọrọ ndị nwere ike ịdị elu.
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 anya
Nlereanya a na-ahụkarị na-aghọ nhọrọ ọkọlọtọ n'akụkụ top-p na top-k na nchịkọta ntinye isi mmalite dị ka llama.cpp na ịmakụ ihu Transformers. Na-atụ anya ka ojiji na-eto eto na-emepụta ihe, mkparịta ụka, na ọgbọ akụkọ ebe ngbanwe nke enweghị nchekwa na-emerụ mma. Ndị na-eme nchọpụta na-agwakọta ya na ọnụ ụzọ na-agbanwe agbanwe nke na-agbanwe n'ọnọdụ ọ bụla yana ijikọta ya na ntaramahụhụ ugboro ugboro. Ka ozi-usoro nzizi ozi na-eto, ụdị nlele anya nwere ike ịgwa akpaka, ụzọ mara nkesa nke ga-ala ezumike nka ntọala ọnọdụ okpomọkụ ejiri aka gee ntị.
Mmejuputa n'ezie n'ụwa
Ịmepụta akụkọ ifo ma ọ bụ uri ebe anyaukwu decoding na-emepụta ihe na-adịghị mma, na-agụgharị akwụkwọ na ndị edemede chọrọ ụdị ọdịdị dị iche iche.
Ịkwado chatbot na-aza na-ezere roboti, nkebi okwu n'ụdị mgbe ị na-adakọ ọnụ yana na isiokwu.
Dị ka ọkọlọtọ ngbanwe (typical_p) na ịmakụ ihu Transformers maka ndị mmepe na-emegharị mmepụta ụdị isi mmalite.
A na-eji ya na oge LLM mpaghara dị ka llama.cpp na text-generation-webui dị ka ihe ọzọ na top-p maka ederede na-adịchaghị njọ.
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
Kọwaa usoro mmepụta, ụda, na ụkpụrụ ịdịmma tupu ibugharị.
Weghachite nzaghachi site na isi mmalite ntụkwasị obi mgbe ọ bụla izi ezi dị mkpa.
Debe ebe nleba anya mmadụ maka mpụta dị elu.
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 Typical Sampling 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
Ntuziaka na-esote
Nnwale nlele anya
Ajụjụ a na-ajụkarị
What is Typical Sampling?
Nlereanya a na-ahụkarị bụ usoro ọgbọ nke ederede na-ewere okwu na-esote site na tokens nke ọdịnaya ọdịnaya ha na-anọdụ ala na ihe ịtụnanya a tụrụ anya ka ihe nlereanya ahụ, kama ijide ndị kacha nwee ike mgbe niile. Ọ na-achọsi ike maka mmepụta nke na-enwe mmetụta nke okike na nke mmadụ site n'ikwekọ ka ezigbo asụsụ si edozi amụma na ihe ọhụrụ.
Kedu isi ihe nlere anya na-atụnyere token onye ọ bụla megide?
Nlereanya a na-ahụkarị na-atụ etu ihe ịtụnanya nke ọ bụla si dị na entropy nkesa, ọdịnaya ozi a na-atụ anya ya, na-edobe akara nke ihe ịtụnanya ya kacha nso na nkezi ahụ.
Kedu ka nlele anya si emeso otu akara kacha nwee ike ime?
Akara nke puru omume nke ukwuu nwere ihe ịtụnanya dị oke ala, nke nwere ike ịdị anya site na nkezi, yabụ ụdị nlele nwere ike hapụ ya n'aka onye ndoro-ndoro ochichi nke edobere maka akara ngosi 'ahụkarị'.
Kedu echiche gbasara asụsụ mmadụ kpalitere ụdị nlele anya?
Usoro a na-adọta n'echiche bụ na asụsụ okike na-anọ nso n'otu ọnụego ozi, yabụ na ọ na-akwado akara ngosi na-ebu ihe ịtụnanya dị nso na nkezi.
Kedu ihe paramita tau na njikwa nlele anya?
Tau na-esetịpụ ihe gbasara puru omume mkpokọta; Tokens kacha nso na entropy na-agbakwunye ruo mgbe ha jikọtara ihe gbasara nke puru omume ruru tau, na-akọwa ihe set ka sample si.
Kedu ụdị sọftụwia a na-ahụkarị maka ndị nrụpụta?
A na-emejuputa nleba anya dị ka nhọrọ ngbanwe (dịka ọmụmaatụ ụdị_p) n'ime ọba akwụkwọ dị ka ịmakụ ihu Transformers na oge ọsọ mpaghara dị ka llama.cpp.