- I-Mu iyimodeli yolimi olusha lwe-Microsoft, elungiselelwe ukusebenza endaweni Windows 11 amadivayisi anama-NPU.
- Ukuhlanganiswa kwayo kokuqala kwenziwa ku-ejenti yokulungiselela ye-Windows 11, evumela ukulungiswa kusetshenziswa ulimi lwemvelo.
- I-Mu igqamile ngokusebenza kwayo kanye nesivinini, ifinyelela ngaphezu kwamathokheni ayi-100 ngomzuzwana ngenxa yemingcele yayo eyizigidi ezingama-330.
- Ihlanganisa izinto ezintsha ezifana ne-Dual LayerNorm, i-RoPE, ne-GQA, futhi iqeqeshwe kusetshenziswa izinqubo ezithuthukisiwe kanye nedatha yemfundo yezinga eliphezulu.
Ukufika kuka Mu, imodeli yakamuva yolimi oluncane eyethulwe ngu I-Microsoft, iphawula isinyathelo esibalulekile kuthrendi yamanje yokubeka ubuhlakani bokwenziwa ngokuqondile kumadivayisi wabasebenzisi. Ngenhloso ye nciphisa ukuncika kwamafu futhi usebenzise amandla we I-Neural Processing Units (NPU), Mu ihlanganiswe ku PCs Copilot+ ejecutando I-Windows 11, ekuqaleni egxile ku Uhlelo lokusebenza lwezilungiselelo ukwenza lula ukufinyelela nokuguqulwa kwemingcele yesistimu kusetshenziswa ulimi lwemvelo nje.
Lokhu kusengaphambili kusho ukuthi, esikhundleni sokuthumela imibuzo kumaseva angaphandle, ukucubungula nezimpendulo kukhiqizwa edivayisini ngokwayo, iqinisekisa ubumfihlo obuningi, ukushesha nokusebenza kahle. Okwamanje, i- Ukukhishwa kuqondiswe kubahlanganyeli bohlelo lwe-Windows Insider ngamakhompyutha we-Copilot+., nakuba okulindelekile ukuthi lobu buchwepheshe buzonwetshwa kubasebenzisi abaningi kanye nemisebenzi kuzibuyekezo ezizayo.
Uyini ngempela u-Mu futhi yini eyenza igqame?
Mu kuyi- imodeli yolimi oluncane (SLM, por sus siglas en inglés), baqeqeshwe ngamapharamitha ayizigidi ezingama-330Usayizi wayo ohlangene akusho ukuzidela ekusebenzeni, njengoba ngokusho kweMicrosoft ifinyelela izibalo ezisondele kakhulu kumamodeli amakhulu kakhulu njenge Phip-3.5-miniLe bhalansi ifinyelelwe ngenxa yenqubo yokuqeqesha eqinile efake amasu afana nalokhu Dual LayerNorm, Rotary Positional Embeddings (RoPE) y Grouped-Query Attention (GQA) enikeza ukusebenza kahle nokunemba, ikakhulukazi kumadivayisi anezinsiza ezilinganiselwe.
Imodeli isebenzisa ngokunenzuzo a isakhiwo se-encoder-decoder yohlobo lwe-transformer, ekwazi ukucubungula okokufaka komsebenzisi nokuyiguqula ibe yizenzo ngaphakathi kwesistimu. Ngenxa yalesi sakhiwo, uM ihlukanisa ukucubungula okokufaka nokuphumayo, okuyinto kunciphisa ukubambezeleka nokusetshenziswa kwenkumbulo, amaphuzu abalulekile wokuqinisekisa ulwazi lomsebenzisi olushelelayo, olungalindi.
Ekuhlolweni okusemthethweni nedatha, u-Mu ufakazele ukuthi uyakwazi phendula amathokheni angaphezu kuka-100 ngomzuzwana futhi unikeze izimpendulo ngaphansi kwama-millisecond angu-500Lezi zinombolo zivumela ukusebenzisana cishe ngokushesha, ngisho noma kuziwa ekulungiseni izilungiselelo noma ukuhumusha imibuzo emide nehlukahlukene ngolimi lwansuku zonke. Uma ungathanda ukumba ujule ekutheni la mamodeli asebenza kanjani, ungabheka Ukuqhathanisa phakathi kwamamodeli olimi ku-PC.
Ukuhlanganiswa Kumenzeli Wokucupha nemisebenzi ebonakalayo
Ukufika kokuqala kuka-Mu kugxile ku- Windows 11 Umenzeli Wokulungiselela, isici esivumela abasebenzisi lungisa amapharamitha wesistimu ngokumane uthayiphe noma usho abakudingayo. Ngokwesibonelo, vele ubuze "Ngiyenza kanjani imodi emnyama isebenze?" o "Ngifuna ukwandisa ukukhanya" ukuze u-Mu akwazi ukuhumusha lowo myalelo esenzweni sobuchwepheshe esihambisanayo ngaphakathi kwesistimu.
IMicrosoft igcizelele ukuthi i-AI ijwayelana nayo amashumi ezinkulungwane zezimo nemibuzo eyahlukeneEqinisweni, sekusetshenziswe ezingaphezu kuka-100,000. 3,6 million amasampula okuqeqeshwa ukumboza yonke into kusukela kuzicelo ezivame kakhulu - njengokushintsha ulimi noma ukuphatha amanethiwekhi e-Wi-Fi - kuya emisebenzini eyinkimbinkimbi. Ngemibuzo emifushane kakhulu noma engaqondakali, isistimu isebenzisa i imisebenzi yokusesha yendabuko, kodwa lapho imiyalelo icacile futhi inemininingwane, i-Mu yenza ngokuzenzakalelayo noma iqondise umsebenzisi isinyathelo ngesinyathelo.
Ubuchwepheshe nokwenza kahle kujwayelane nezizukulwane ezintsha zehadiwe
La Mu optimization kube ngelinye lamaphuzu acatshangelwe ngokucophelela kakhulu ngesikhathi sokuthuthukiswa kwawo. I-Microsoft isebenze ngokubambisana nabalingani be-silicon njenge I-AMD, i-Intel ne-Qualcomm ukuze uyivumelanise nemininingwane yama-NPU amasha akhona kuma-Copilot+ PCLo msebenzi ohlanganyelwe wenze kwaba nokwenzeka ukwethula amasu okulinganisa emva kokuqeqeshwa, okuguqula izisindo zemodeli nokwenza kusebenze kube izinombolo eziphelele ezingu-8- kanye ne-16-bit, ngaleyo ndlela kunciphise ukusetshenziswa kwememori nokugwema isidingo sokuqeqesha kabusha yonke imodeli.
Inqubo yokuqeqeshwa kukaMu yenziwa ezindaweni ezisebenza kahle kakhulu, kusetshenziswa I-NVIDIA A100 GPUs ngaphakathi Azure Machine LearningIsethi yedatha ifakiwe amakhulu ezigidigidi zamathokheni okufundisa kanye namasu anjengokuthi i-distillation evela kumamodeli we-Phi kanye ne-Low-Range Adaptation (LoRA) ukudlulisa ulwazi nokushuna kahle imodeli emisebenzini ethile. Umphumela uyimodeli encane, eguquguqukayo eguqulelwe ngokukhethekile ezinsizeni kanye nemikhawulo yehadiwe ephathekayo yesimanje. Ungahlola futhi ukuthi kanjani vula i-PC yakho ibe ihabhu le-AI lendawo ukwandisa amakhono esistimu yakho.
Izinselele zamanje, ukutholakala kanye namathemba esikhathi esizayo
Enye yezinselelo ezinkulu ezibhekene noMu yi ukuchazwa kwemibuzo engacacile noma emifushane kakhulu, inkinga evamile ezinhlelweni ezisekelwe olimini zemvelo. Ukwenza lokhu, IMicrosoft isebenzise i-hybrid logicNakuba imibuzo emifushane ibangela imiphumela yosesho evamile, imiyalelo enemininingwane eminingi ibangela ukungenelela kwe-AI, noma ukuqondisa umsebenzisi noma ukwenza izenzo ezizenzakalelayo.
Okwamanje, I-Mu itholakala kuphela ngesiNgisi nakumadivayisi e-Copilot+ ngesiteshi i-Insider., nakuba kulindeleke ukuthi inwetshelwe ezilimini eziningi namanye amadivaysi ezinyangeni ezizayo, okuhlanganisa lawo anama-AMD nama-Intel processors. privacidad y la seguridad Ziphinde zidlale indima ebalulekile, uma kubhekwa isimo sendawo sokucutshungulwa.
Ukuthunyelwa kwe-Mu kumane kuyisiqalo sesu elibanzi le-Microsoft lokuhlanganisa I-AI yasendaweni namamodeli olimi asebenzayo ezinhlelweni eziningi nakakhulu nezici zesistimu yokusebenza, ukuthuthukisa ulwazi nokufinyeleleka ngaphandle kokudela ukusebenza noma ubumfihlo.
Ngingumshisekeli wezobuchwepheshe oguqule izintshisekelo zakhe "ze-geek" zaba umsebenzi. Ngichithe iminyaka engaphezu kwengu-10 yempilo yami ngisebenzisa ubuchwepheshe obusezingeni eliphezulu kanye nokukitaza ngazo zonke izinhlobo zezinhlelo ngenxa yelukuluku lokufuna ukwazi. Manje sengiqeqeshelwe ubuchwepheshe be-computer nemidlalo yama-video. Lokhu kungenxa yokuthi sekuphele iminyaka engaphezu kwengu-5 ngisebenza ngokubhalela amawebhusayithi ahlukahlukene ezobuchwepheshe nemidlalo yevidiyo, ngenza izindatshana ezifuna ukukunikeza imininingwane oyidingayo ngolimi oluqondakala yiwo wonke umuntu.
Uma unemibuzo, ulwazi lwami lusukela kuyo yonke into ehlobene nesistimu yokusebenza ye-Windows kanye ne-Android yomakhalekhukhwini. Futhi ukuzibophezela kwami kuwe, ngihlala ngizimisele ukuchitha amaminithi ambalwa futhi ngikusize uxazulule noma yimiphi imibuzo ongase ube nayo kulo mhlaba we-inthanethi.