Hy Memory 2.0 Uyasebenza

I-Hy Memory 2.0 iwuhlaka lwememori lwama-agent olukhishwe yi-Tencent Hunyuan njengomthombo ovulekile, olusebenzisa ukwakheka kwezingqimba ezintathu: imodeli + uhlaka + uhlelo lokusebenza: imodeli i-Hy-Memory-30A3B eqeqeshelwe inkumbulo, i-SDK esekela izinhlobo ezintathu zezinkumbulo, okuyingxoxo, umsebenzi kanye nekhodi, kanye nohlelo lokusebenza lokuphatha inkumbulo lwasendaweni i-MemHub olungaxhuma kumathuluzi e-AI angu-10. I-Hy Memory 2.0 ingagcina futhi ivumelanise izinkumbulo zomuntu siqu phakathi kwezingxoxo namathuluzi ahlukahlukene, inikeze i-Agent inkumbulo yesikhathi eside eqhubekayo nephelele. Isivele isetshenziswa ezinsizeni ezifana ne-Tencent Stock Picks ne-Tencent News.

Ukwakhiwa kwezingqimba ezintathu okukhishwe ngokuphelele njengomthombo ovulekile: imodeli ekhethekile, uhlaka lwememori kanye nohlelo lwe-MemHub kungasetshenziswa ngokuzimela noma kuhlanganiswe kube yisixazululo esiphelele se-Memory.
Ukuphathwa kwezinhlobo ezintathu zezinkumbulo: kugcina, kubuyekeze futhi kuthuthukise izinkumbulo zengxoxo, umsebenzi nekhodi; abathuthukisi bangaxhuma i-Agent nge-SDK.
Inkumbulo yengxoxo: iguqula amaqiniso ahlakazekile abe ukulandelana okuphelele kwezehlakalo, ixhumanise ngokuzenzakalelayo okuqukethwe okuvela ezikhathini nasezingxoxweni ezahlukene futhi iqede ukungqubuzana.
Inkumbulo yomsebenzi: iqopha ngokuningiliziwe izinhloso zomsebenzi, izindlela nemininingwane engasetshenziswa kabusha; ngaphambi kokuyisebenzisa kabusha ihlola izimo ezifanele, isize ekwenzeni imisebenzi yesikhathi eside.
Inkumbulo yekhodi: igcina umongo wamaphrojekthi ezimweni zekhodi, ithuthukise ukunemba kwezimpendulo zemibuzo yekhodi nokuxazululwa kwemisebenzi.
Ukuvumelanisa inkumbulo ye-MemHub: uhlelo lwasendaweni luxhuma kumathuluzi e-AI angu-10 afana ne-Cursor ne-Codex, lungenise izingxoxo zangaphambilini futhi luvumelanise inkumbulo ngesikhathi sangempela phakathi kwamathuluzi.

Landela i-WeChat bese uphendula ngokuthi “开源” ukuze ujoyine iqembu lokuxoxa ngamaphrojekthi e-AI omthombo ovulekile
Ukuhlelwa kwenkumbulo ngokwezehlakalo: kuhlelwa isikhathi, indawo, abantu nezinye izici ngokusekelwe ezehlakalweni, bese kuxhunyaniswa nentuthuko elandelayo; lapho kuphendulwa imibuzo enemikhondo eminingi, imininingwane ephindaphindwayo ihlanganiswa kuqala, okuqukethwe okuphikisanayo kususwe, bese imemori ehlanzekile idluliselwa kumodeli.
Ukusetshenziswa kabusha kolwazi ngokwezimo: izinkumbulo zomsebenzi nezekhodi zigcinwa ngezindlela eziningi, okuhlanganisa inhloso yomsebenzi, indlela esiqinisekisiwe, imiphumela yesigaba kanye nezimo lapho ulwazi lusebenza khona, futhi zihlelwe futhi zibuyekezwe ngokuhamba kwesikhathi; lapho kusetshenziswa ulwazi kabusha, kuqala kuhlolwa ukuthi isimo samanje siyahlangabezana yini nezimo zokusebenza, ukuze kusetshenziswe ngokunembile futhi kugwenywe ukudluliselwa okungalungile.
Ukuqhubeka nomsebenzi endaweni owawumiswe kuyo: lapho umsebenzi wesikhathi eside uqala kabusha, uhlaka luqala ngokunquma udaba olusingathwayo, bese lubuyisa ngokunembile izinqumo ezifanele, ulwazi olusasebenza kanye nenqubo yomsebenzi wangaphambilini, luthole inkumbulo efanele kakhulu kusetshenziswa ama-Token ambalwa.
Imodeli encane ekhethekile: i-Hy-Memory-30A3B iqeqeshelwe ngokukhethekile ukubuyekeza nokuphatha inkumbulo; isebenza kangcono kune-Opus5 ku-Hy Memory, Mem0, MemOS nakwezinye izinhlaka.
Ukuhlanganiswa kwenkumbulo yomhlaba wonke: i-MemHub ine-Hy Memory SDK eyakhelwe ngaphakathi, egcina izingxoxo zawo wonke amathuluzi e-AI endaweni njengezinkumbulo zomuntu siqu ezihlangene, bese ingeza umongo emuva kumathuluzi ngendlela yama-plugin, ihlukanise izimpahla zememori namathuluzi.

Thola izinsiza zomthombo ovulekile: landa uhlaka, uhlelo lokusebenza nezisindo zemodeli ku-GitHub (Tencent-Hunyuan/Hy-Memory, Tencent-Hunyuan/MemHub) naku-HuggingFace (Hy-Memory-30B-A3B).
Abathuthukisi baxhume i-SDK: hlanganisa i-Hy Memory SDK ne-Agent yakho bese ubiza amandla ayo okugcina, okubuyekeza nawokubuyisa inkumbulo.
Abasebenzisi abajwayelekile bafake i-MemHub: landa i-CLI noma i-APP ye-Mac ye-MemHub ukuze uyisebenzise njengesikhungo sendawo sokuphatha inkumbulo.
Gunyaza futhi uxhume amathuluzi e-AI: ku-MemHub, gunyaza amathuluzi e-AI angu-10 asekelwayo, afana ne-WorkBuddy, Cursor ne-Codex.
Ngenisa izingxoxo zangaphambilini: ngemva kokugunyaza, ngenisa ngokuchofoza okukodwa izingxoxo zangaphambilini zawo wonke amathuluzi; i-Hy Memory SDK eyakhelwe ngaphakathi izozigcina njengezinkumbulo zomuntu siqu zasendaweni.
Vula ukuvumelanisa kwesikhathi sangempela: izingxoxo ezintsha ezikhiqizwa amathuluzi ahlukahlukene zizofakwa ngokuzenzakalelayo enkumbulweni yomhlaba wonke efanayo, ngaphandle kokusebenza ngesandla.
Sebenzisa ngendlela ye-plugin: lapho uqala umsebenzi kunoma yiliphi ithuluzi elixhunyiwe, i-MemHub izofaka ngokuzenzakalelayo izinkumbulo ezifanele njengomongo walelo thuluzi.

Izingqimba ezintathu zonke zingumthombo ovulekile: imodeli, uhlaka nohlelo lokusebenza konke kukhishwe njengomthombo ovulekile; kungasetshenziswa ngokuzimela noma kuhlanganiswe, kube nokuguquguquka okuhamba phambili embonini.
Indawo yokuqala kuzo zonke izivivinyo: ezilinganisweni ezi-4 ze-Benchmark ezimweni zengxoxo, umsebenzi nekhodi, iyinombolo yokuqala phakathi kwezinhlaka zememori ezinomthombo ovulekile.
Ukuqanjwa kokuqala embonini: uhlaka lokuqala lwememori oluphakamisa inkumbulo yomsebenzi nenkumbulo yekhodi, luhlanganise izimo eziphelele kakhulu zememori.
Imodeli encane idlula imodeli enkulu: imodeli ekhethekile enama-parameter acishe abe ngu-3B asebenzayo, enomphumela wememori odlula i-Opus5, izindleko eziphansi nokusebenziseka kalula ekusetshenzisweni.
Ukusetshenziswa kwesitoreji kuncane kakhulu: nakuba inemiphumela ehamba phambili kuzo zontathu izimo, ukusetshenziswa kwendawo yesitoreji sememori kuncane kakhulu, ilungele ukusetshenziswa isikhathi eside.
Ukusebenza kahle kwama-Token kuvelele: ku-LoCoMo, ukusetshenziswa kwama-Token kwehle ngo-39.3%, kutholwe ukunemba okuphezulu ngomongo omncane.
Izimpahla zememori ziba yimpahla esebenzisekayo: i-MemHub ixhumanisa amathuluzi e-AI angu-10, yenze inkumbulo yomuntu siqu iqhubeke phakathi kwamathuluzi futhi ingaboshelwa kumkhiqizo owodwa.

Indawo yokugcina ye-GitHub: https://github.com/Tencent-Hunyuan/MemHub
https://github.com/Tencent-Hunyuan/Hy-Memory
Indawo yamamodeli ye-HuggingFace: https://huggingface.co/tencent/Hy-Memory-30B-A3B

Umsizi wocwaningo lokutshalwa kwezimali kwezezimali: ukhumbula izigaba nezinqubo zokuhlaziya ezithandwa umsebenzisi ezinsukwini nasezingxoxweni ezahlukene; izinga lokubuyisa izimpendulo ezinezinyathelo eziningi ku-Stock Picks likhuphuke lisuka ku-50% laya ku-90%.
Ukuncoma okuqukethwe kwezindaba: ukhumbula izihloko ezithandwa umsebenzisi nendlela athanda ukubizwa ngayo; ku-“智声” naku-“新闻妹” ze-Tencent News, isilinganiso sokuhlolwa kokuthandwa kwesikhathi eside sifinyelele ku-91.68.
Imibuzo nezimpendulo zokuphathwa kolwazi: i-iWiki iphendula isebenzisa inkumbulo ebuyisiwe kuphela, futhi ukunemba kwayo kuphakeme ngo-48.6% uma kuqhathaniswa nokunikezwa zonke izingxoxo zangempela.
Usizo lokuthuthukiswa kohlelo: ikhumbula izindinganiso zephrojekthi nokuthandwa komuntu siqu kokubhala ikhodi phakathi kwamathuluzi, iphinde isebenzise inkumbulo yekhodi ngaphandle komthungo ku-Cursor, Codex nakwamanye amathuluzi.
Ihhovisi nemisebenzi yesikhathi eside: iqopha izinhloso zomsebenzi, imiphumela yezigaba nolwazi olungasetshenziswa kabusha, isekele i-Agent yasehhovisi ekuqhubekeni nomsebenzi lapho izingxoxo eziningi ziqala futhi zimiswa.