Ntụle echiche nke osisi
Osisi-echiche na-ahapụ ihe nlereanya inyocha ọtụtụ ụzọ ntụgharị uche n'otu aka ahụ, dị ka alaka osisi, kama itinye aka n'otu ahịrị echiche.
Nchịkọta
It can look ahead, evaluate partial solutions, and backtrack from dead ends.
Ime miri emi
Yao et al webatara. N'afọ 2023, Tree-of-Thoughts (ToT) na-achịkọta mkpali echiche n'usoro. Ebe echiche-iche na-emepụta otu usoro usoro ntụgharị uche, ToT na-ahazi nsogbu dị ka osisi: ọnụ nke ọ bụla bụ ihe ngwọta nke akụkụ ('echiche'), ihe nlereanya ahụ na-emepụta ọtụtụ echiche na-esote echiche site na ọnụ nke ọ bụla. Nzọụkwụ nleba anya dị iche na-egosi etu alaka ọ bụla si na-ekwe nkwa, na algọridim nchọta dị ka nchọta obosara-mbụ ma ọ bụ omimi-mbụ na-ekpebi alaka ndị a ga-agbasa na nke a ga-akwacha. Nke a na-eme ka ihe nlereanya ahụ kpachara anya nyochaa, leba anya n'ihu nzọụkwụ ole na ole, ma laghachi azụ mgbe ụzọ na-adịghị mma. ToT na-enwu n'ọrụ ndị na-emeri echiche otu ụzọ anyaukwu, nke kachasị ama egwuregwu nke 24, ebe GPT-4 nwere agbụ-echiche na-edozi ihe dịka 4% nke mgbagwoju anya mana ToT kpaliri ọganihu ruo ihe dịka 74%.
Nghọta nka nka
ToT nwere iberibe atọ: onye na-emepụta echiche nke na-atụ aro usoro ndị ọzọ ga-esote, onye nyocha steeti (na-abụkarị otu LLM ahụ kpaliri ọnụ ma ọ bụ votu na ngwọta akụkụ dị ka 'n'ezie / ma eleghị anya / agaghị ekwe omume'), na usoro ọchụchọ (BFS, DFS, ma ọ bụ nchọgharị beam) nke na-agagharị n'osisi ahụ. N'ihi na ihe nlereanya a na-enyocha ele mmadụ anya n'ihu steeti na prunes adịghị ike alaka, ọ allocates gbakọọ n'ebe na-ekwe nkwa na mpaghara nke ngwọta ohere, trading extra inference oku maka substantially elu izi ezi na siri ike nsogbu.
Mmetụta atụmatụ
Ọnụ ego na mmefu ego
Mkpebi ihe owuwu ụlọ na-akwalite arụmọrụ yana ọnụ ahịa ọrụ ruo ọtụtụ afọ.
Mkpebi doro anya
Nkà mmụta nka na-enyere ndị otu egwuregwu aka ịhọrọ nchịkọta ziri ezi, ọ bụghị naanị nke kachasị ọhụrụ.
Quality akara
Nhọrọ injinia ka mma na-ebelata ihe omume ntụkwasị obi na mmepụta.
Ọdịnihu nke Ntụgharị-echiche nke Osisi
Osisi-nke echiche na-emetụta ka esi etinye oge ngụkọ: kama otu nnukwu ngafe na-aga n'ihu, sistemụ na-achọwanye ụzọ ntụgharị uche wee họrọ nke kacha mma. Ọrụ n'ọdịnihu na-achọ ịmụta mgbe alaka ụlọ ọrụ bara uru ọnụ ahịa ya dị elu, iji ọrụ bara uru a zụrụ azụ kama nyochaa onwe ya ngwa ngwa, na ịtụgharị osisi n'ime usoro ntụgharị uche ka ntule ahụ wee rụọ ọrụ nke ọma. Na-atụ anya mwekota siri ike na ngwaọrụ na ndị na-enyocha ya ka nlebanya alaka ụlọ ọrụ dabere na nzaghachi n'ezie kama ịkọ nkọ ihe nlereanya ahụ.
Mmejuputa n'ezie n'ụwa
Ịdozi Egwuregwu nke 24 site n'ịkọba nọmba abụọ iji jikọta na mbụ na ịkwachaa ụzọ mgbakọ na mwepụ na-enweghị ike iru 24.
Edemede ihe okike nwere atụmatụ jikọrọ ọnụ, ebe ihe nlereanya ahụ depụtara ọtụtụ atụmatụ atụmatụ, na-enyocha ha, ma gbasaa nke kachasị ike tupu ya edepụta prose.
Mgbochi mgbagwoju anya dị ka obere okwu mkparịta ụka, ebe okwu ọ bụla jupụtara bụ echiche ma hapụ alaka na-adabaghị adaba.
Nsogbu mgbakọ na mwepụ ọtụtụ nzọụkwụ ma ọ bụ atụmatụ nhazi ebe ihe nlereanya na-enyocha usoro dị n'etiti ọzọ na azụ azụ site na ndị na-emebi mmachi.
Ihe ize ndụ & okporo ụzọ nche
Ịkwalite otu akara ngosi nwere ike zoo adịghị ike sistemụ sara mbara.
A na-eledakarị ihe akụrụngwa na ụgwọ ọrụ anya.
Ọdịiche nchekwa na nleba anya nwere ike itolite ka sistemu na-adịwanye mgbagwoju anya.
Map mmejuputa
Kọwaa latency, ịdịmma na ebumnuche ọnụ ahịa tupu mmejuputa ya.
Benchmark n'okpuru ibu dị adị na ọnọdụ data.
Nleba anya akụrụngwa maka mperi, ịkpafu na mmetụta onye ọrụ.
Kwadebe ụzọ nzaghachi azụghachi azụ na ihe omume tupu ịchachaa.
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 Tree-of-Thoughts Reasoning 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
Osisi Echiche
Ajụjụ a na-ajụkarị
What is Tree-of-Thoughts Reasoning?
Osisi-echiche na-ahapụ ihe nlereanya inyocha ọtụtụ ụzọ ntụgharị uche n'otu aka ahụ, dị ka alaka osisi, kama itinye aka n'otu ahịrị echiche. Ọ nwere ike na-ele anya n'ihu, nyochaa ihe ngwọta ụfọdụ, na azụ azụ site na njedebe nwụrụ anwụ.
Kedu ka Osisi-nke echiche si dị iche na mkpali-ntụgharị uche?
Chain-of-thought na-emepụta otu usoro ahịrị ahịrị, ebe Osisi-echiche na-alaka n'ime ọtụtụ ụzọ ndị ndoro-ndoro anya nke enwere ike ịtụle na kwachaa.
Na Tree-of-Thoughts, gịnị ka otu ọnụ na-anọchi anya ya?
Ọnụ ọ bụla bụ ihe ngwọta nke akụkụ (echiche), na ihe nlereanya na-agbasa ọtụtụ echiche na-esote echiche site na ọnụ ọ bụla.
Kedu ọrụ onye nyocha steeti na-arụ na Osisi-nke echiche?
Onye nleba anya, na-abụkarị otu LLM, ọnụego ma ọ bụ votu na mpaghara akụkụ (dịka ọmụmaatụ, ijide n'aka / ikekwe / agaghị ekwe omume) yabụ ọchụchọ ahụ nwere ike ilekwasị anya na ngalaba na-ekwe nkwa.
N'egwuregwu nke 24 benchmark, ToT pụtara n'ike n'ike n'ike n'ike n'ike n'ike n'ike site n'ịwụ elu?
GPT-4 jiri yinye-echiche dozie ihe dị ka 4% nke egwuregwu egwuregwu egwuregwu 24, ebe echiche nke osisi ruru 74%.
Kedu usoro ọchụchọ a na-ejikarị eme njem n'osisi echiche?
ToT ụzọ abụọ chere ọgbọ na nlebanya na usoro nchọcha dị ka BFS, DFS, ma ọ bụ nchọ beam iji kpebie alaka ndị ga-agbasa.