Kwenzekeni
Abacwaningi bethule i-AutoViewMem, uhlaka olusha oluklanyelwe ukuthuthukisa inkumbulo yesikhathi eside kumamodeli amakhulu wolimi (LLM). Ngokususa umthwalo wenhlangano ye-semantic kusukela esikhathini sokuthola ukuze ubhale, isistimu idala ukubukwa okuzilungiselelayo, ukunqwabelana okuphansi kwedatha yengxoxo. Le ndlela ihlose ukuxazulula inkinga yokuphazanyiswa kwe-semantic, lapho ulwazi oluhlukile-njengokuthandwa ngabasebenzisi, izehlakalo ezithile, kanye nezingqinamba zesikhashana-luxubene nezinhlelo zenkumbulo zendabuko, ezimele okukodwa.
I-AutoViewMem isebenza ngokuthola ukubukwa kwenkumbulo yekhandidethi ekulandeleleni ukuxhumana nokukhetha isethi yokubuka ehlangene, ehambisanayo. Esikhundleni sokugcina lonke ulwazi ekumeleleni okukodwa, okuxubile, uhlaka lusebenzisa le mibono ukuze iqondise ukukhishwa okuhlelekile kwezinkumbulo ngesikhathi ezibhalwa ngaso.
Uhlaka lusebenzisa isinyathelo sokuhlanganisa ungaxhunyiwe ku-inthanethi ukuze kuqinisekiswe ukubumbana kwenkumbulo nokuvumelana, okusiza ukunciphisa ukuqoqwa kolwazi olungafuneki noma olungqubuzanayo ngokuhamba kwesikhathi.
Ekuhloleni, abacwaningi basebenzise amamodeli we-Qwen3-8B kanye ne-Qwen3-14B. Imiphumela ibonise ukuthi i-AutoViewMem isebenze kahle kakhulu kunezisekelo zememori ezikhona ekuphenduleni imibuzo emkhathizwe ende nemisebenzi eqondene nawe ngenkathi igcina ipayipi elivamile, elilula lokukhomba.
Imininingwane yomthombo: arxiv.org ↗
Kungani kubalulekile
I-AutoViewMem ibhekana nengqinamba ebalulekile ekuthuthukisweni komenzeli we-AI: ukungakwazi ukukhumbula ngendlela enokwethenjelwa kanye nokuhlanganisa ulwazi ngokusebenzisana okunwetshiwe. Ngokuhlukanisa inkumbulo endaweni yokugcina, uhlaka luvumela izindlela zokubuyisa ezijwayelekile ukuthi zisebenze ngempumelelo ngaphandle kokudinga umzila onzima, obiza ngokwekhompyutha noma izinqubo zokusesha eziphindaphindwayo. Lokhu kuthuthukiswa kokunemba kwenkumbulo kuthinta ngokuqondile ukwethembeka kwama-ejenti e-AI emisebenzini esemkhathizwe ende, njengokugcina abantu abangaguquki babasebenzisi noma ukulandelela izithiyo zephrojekthi eziyinkimbinkimbi ngokuhamba kwesikhathi. Abacwaningi babonise izinzuzo zokusebenza kumabhentshimakhi we-LoCoMo kanye ne-PersonaMem besebenzisa amamodeli e-Qwen3-8B kanye ne-Qwen3-14B, bephakamisa ukuthi lolu shintsho lwezakhiwo lunganikeza usizo olubalulekile konjiniyela abakha izinhlelo zokusebenza ze-AI eziqhubekayo, eziqinile.
Amasistimu amanje enkumbulo ngokuvamile ahlushwa 'ukuphazamiseka kwe-semantic,' lapho ukubuyiswa kolwazi oluhlobene kuvinjwa umsindo obangelwa ukuxutshwa kwezinhlobo ezahlukene zedatha (isb., amaqiniso ngokumelene nokuthandwayo). Idizayini yokumela yokuqala ye-AutoViewMem ihlukanisa ngempumelelo lokhu kukhathazeka ngaphambi kokuba idatha ikhonjwe.
Ngokuhambisa inqubo yokuhlukanisa ukuze ibhale isikhathi, uhlaka lugwema isidingo sezakhiwo eziyinkimbinkimbi, ezinezinyathelo eziningi zokubuyisa, okwenza kube lula ukulisebenzisa ngaphakathi kwamapayipi e-ejenti ye-AI ekhona.
Ikhono lokugcina inkumbulo enembile, yesikhathi eside liyimfuneko ebalulekile kubasebenzeli abahloselwe ukusebenza njengabasizi bomuntu siqu noma abahlanganyeli besikhathi eside, njengoba lithonya ngokuqondile ikhono le-ejenti lokuhlala lingashintshile futhi liwazi umongo phakathi namaviki noma izinyanga zokusebenzisana.
I-Interactive Mechanism: Indlela Esebenza Ngayo Ngempela
Hlola ubuchwepheshe obuyisisekelo ngemuva kwalokhu kuthuthukiswa ngokuhlanganyela.
crm_get_transaction(id='4092').An agent must create a draft calendar event for Tuesday at 2 p.m. Which evidence would establish the requested result?
Ongakubuka ngokulandelayo
Okuyinhloko okungaziwa ukuthi i-AutoViewMem ikala kanjani ezindaweni zokukhiqiza ezinamadathasethi amakhulu kakhulu noma izinhlobo eziningi zokusebenzisana ezihlukene kunalezo ezihlolwe kumabhentshimakhi we-LoCoMo kanye ne-PersonaMem. Nakuba abacwaningi babika ukusebenza okuthuthukisiwe kumgogodla we-Qwen3, ukusebenza kahle kohlaka kuzo zonke izinhlobo zezakhiwo ezihlukene kanye nomthelela walo wokubambezeleka phakathi nesigaba sokukhipha 'isikhathi sokubhala' kusazobonakala emhlabeni wangempela, ukusetshenziswa kwethrafikhi ephezulu. Izibuyekezo zesikhathi esizayo zingacacisa ingqikithi yokubala yenqubo yokuhlanganisa ungaxhunyiwe ku-inthanethi nokuthi lolu hlaka lungahlanganiswa yini nengqalasizinda yesizindalwazi esikhona ngaphandle kwezinguquko ezibalulekile.
Ucwaningo okwamanje lukhawulelwe kuma-benchmarks athile (i-LoCoMo ne-PersonaMem). Akukacaci ukuthi uhlaka lusingatha kanjani ukusakazwa kolwazi olushintshashintshayo noma olushintsha ngokushesha ezindaweni ezibukhoma, ezinabasebenzisi abaningi.
Izindleko zokubala zesigaba 'sokuhlanganisa ungaxhunyiwe ku-inthanethi' azichazwanga ngokugcwele ngokwezidingo zensiza zokusatshalaliswa okukhulu.
Onjiniyela kufanele baqaphe ukuthi lolu hlaka lwamukelwa yini abahlinzeki besizindalwazi abakhulu be-vector noma luhlanganiswe nezinhlaka ezidumile ze-ejenti, ezingabonisa ukusebenza kwalo okungokoqobo kwezinhlelo zokusebenza zesikali somkhakha.