Ntaramahụhụ ugboro ugboro na njikwa ngbanwe
Njikwa ngbanwe bụ mkpịsị aka na-ekpebi ka ụdị asụsụ si ahọrọ mkpụrụokwu ọ bụla na-esote na nkesa puru omume.
Nchịkọta
Settings like temperature, top-p, and repetition penalty shape whether output feels creative, focused, or stuck in loops.
Ime miri emi
Ụdị asụsụ anaghị ewepụta ederede ozugbo; ọ na-ewepụta ohere maka akara ngosi ọ bụla enwere ike. Idozi koodu bụ atụmatụ maka ịtụgharị ohere ndị ahụ ka ọ bụrụ okwu n'ezie. Okpomọkụ na-emegharị nkesa: ọnụ ala dị ala na-eme ka ọ dịkwuo mma na akara ngosi (elekwasị anya, mkpebi siri ike), ụkpụrụ dị elu na-eme ka ọ dị nro (iche iche, dị ize ndụ). Top-k na-edobe naanị k kacha puru omume token; top-p (nucleus sampling) na-edobe nke kacha nta nke ihe puru omume ya ruru n'ọnụ ụzọ dịka 0.9. Ntaramahụhụ ugboro ugboro na-ekewa ọnụọgụ akara ndị ejirila, na-akụda ihe nlereanya ahụ ikwugharị onwe ya. Njikwa ndị emetụtara gụnyere ntaramahụhụ ugboro ole (gbara ugboro ole akara ngosi pụtara) na ntaramahụhụ ọnụnọ (ntaramahụhụ dị larịị ozugbo akara pụtara ma ọlị). Ntugharị ihe ndị a na-egbochi ma loops rọbọt yana ịgba egwu na-enweghị njikọ.
Nghọta nka nka
Ntaramahụhụ ugboro ugboro na-arụ ọrụ na ọkwa logit. Tupu ịtụgharị akara ka ọ bụrụ ihe puru omume site na softmax, a na-ekewa logit nke akara ngosi ọ bụla ewepụtara na mbụ site na ihe ntaramahụhụ (nke 1.1 ruo 1.3) ma ọ bụrụ na ọ dị mma, ma ọ bụ mụbaa ma ọ bụrụ na ọ dị njọ. Nke a na-ebelata ohere nke ịhọrọgharị akara ndị ahụ. Ntaramahụhụ ugboro ugboro na-ewepụ ego dabara na ọnụ ọgụgụ token, ebe ntaramahụhụ ọnụnọ na-ewepụ ego a kapịrị ọnụ ozugbo akara pụtara, n'agbanyeghị ugboro ole.
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 ntaramahụhụ ikwugharị na njikwa ngbanwe
Ndoputa bụ mpaghara nyocha na-arụsi ọrụ ike. Ụzọ ọhụrụ dị ka nchọta dị iche, nlele ahụkarị, eta-sampling na min-p sampling bu n'obi ịhazi ịdị n'otu na ụdị dị iche iche nke ọma karịa ọnụ ụzọ edoziri. Ndozi nleba anya na-eji obere ihe atụ iji mee ka ọgbọ dị ngwa. Na-atụ anya ka sistemu na-eme n'ọdịnihu na-emegharị paramita ngbanwe n'ụzọ dị iche iche, yana ikpughe njikwa ọkwa dị elu dị mfe ka ndị ọrụ nwee ike ịrịọ 'ihe okike karịa' ma ọ bụ 'kpọmkwem karịa' na-ejighi aka jugging okpomọkụ na ntaramahụhụ.
Mmejuputa n'ezie n'ụwa
Ngwa ọdịde ihe okike na-ebuli okpomoku na elu-p iji mepụta akụkọ dị iche iche na-eju anya.
Onye nnyemaaka koodu na-eweda ọnọdụ okpomọkụ dị nso na efu ka ọ weghachite otu nke kachasị, koodu mmecha deterministic.
A chatbot na-etinye ntaramahụhụ ugboro ugboro n'ihe dị ka 1.2 iji kwụsị ya ịdọrọ otu nkebiokwu ahụ ugboro ugboro.
Onye ọrụ API na-esetịpụ ntaramahụhụ ugboro ole iji kụda onye nchịkọta akụkọ iji otu okwu buzz gabiga ókè n'ofe akwụkwọ dị ogologo.
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 Repetition Penalty and Decoding Controls 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
Modeldị Draft Ndozi Apụrụiche
Ajụjụ a na-ajụkarị
What is Repetition Penalty and Decoding Controls?
Njikwa ngbanwe bụ mkpịsị aka na-ekpebi ka ụdị asụsụ si ahọrọ mkpụrụokwu ọ bụla na-esote na nkesa puru omume. Ntọala dị ka okpomọkụ, top-p, na nkwụghachi ụgwọ udi ma mmepụta nwere mmetụta okike, lekwasịrị anya, ma ọ bụ rapaara na loops.
Kedu ihe iweli oke 'okpomọkụ' na-eme na mmepụta ihe nlereanya?
Okpomọkụ dị elu na-eme ka nkesa nke puru omume, na-eme ka akara ngosi dị ntakịrị karịa asọmpi yana mmepụta dị iche iche na enweghị atụ.
Kedu ka nlele top-p (nucleus) si ekpebi akara ndị a ga-atụle?
Top-p na-ahọpụta otu kacha nta nke akara ngosi kachasị elu nke ihe gbasara puru omume ruru n'ọnụ ụzọ (dịka, 0.9), yabụ ọdọ mmiri a na-eme ntuli aka dabara na okirikiri.
Kedu ọkwa ntaramahụhụ nkwugharị na-emekarị?
Ntaramahụhụ ugboro ugboro na-agbanwe akara ngosi (ọnụọgụ raw) nke akara ejirila tupu agbanwee ka ọ bụrụ ihe puru omume, na-ebelata ohere nhọrọ ha.
Gịnị bụ isi ihe dị iche n'etiti ntaramahụhụ ọnụnọ na ugboro ole?
Ntaramahụhụ ugboro ugboro na-eto n'ọnụọgụ ugboro ejirila akara, ebe ntaramahụhụ ọnụnọ na-emetụta otu mbelata oge a kapịrị ọnụ ozugbo akara pụtara ma ọlị.
Maka onye inyeaka koodu kwesịrị iweghachi otu mmecha a pụrụ ịdabere na ya, kedu ntọala kacha kwesị ekwesị?
Okpomọkụ dị nso na efu na-eme ka nzipụta ihe fọrọ nke nta ka ọ bụrụ ihe ekpebisi ike, ọ fọrọ nke nta ka ọ bụrụ mgbe niile na-ahọpụta akara puru omume kachasị elu, nke dị mma maka koodu enwere ike ịkọ.