- Ukuhlanganiswa okugcwele kwe-Tensor ML SDK ne-LiteRT ukuze kuthuthukiswe ukuthunyelwa kwamamodeli e-AI kumadivayisi e-Pixel.
- Ukufinyelela Engadini Yamamodeli enamamodeli angaphezu kwe-100 alungiselelwe i-Google Tensor TPU.
- Ukusekelwa okuthuthukisiwe kwamamodeli olimi olunezindlela eziningi kanye nombono ngokusebenzisa ukwakheka kwe-Gemma 4.
- Ikhono lokwenza iziphetho zesivinini esiphezulu kanye nobumfihlo ngenxa yehadiwe ye-Tensor G5.
Uma uthanda ukuthuthukiswa kohlelo lokusebenza kanye nobuhlakani bokwenziwa, cishe uke waphawula ukuthi umdlalo uyashintsha. I-Google inqume ukuvula iminyango yehadiwe yayo enamandla kakhulu nge isoftware yokuthuthukisa i-del I-Tensor ML SDK, ithuluzi elivumela abathuthukisi ukuthi bazuze kakhulu kuyunithi yokucubungula i-tensor (TPU) yamadivayisi e-Pixel, okwenza i-AI ingaxhomeki efwini kodwa indize ngqo kudivayisi.
Into ejabulisa kakhulu ukuthi le SDK ishiye isigaba sayo sokuhlola ngemuva futhi yangena Isigaba se-BetaLokhu kusho ukuthi akuseyona eyabambalwa abakhethiwe kuphela; noma yimuphi umhleli wezinhlelo angaqala ukudala okuhlangenwe nakho kwe-AI okusebenzisanayo, okuyimfihlo, futhi, ngaphezu kwakho konke, okusheshayo kakhulu, okusebenzisa ukwakheka kwe-Tensor SoC ye-Google ukwenza imisebenzi eyayibonakala ingenakwenzeka ukuyenza endaweni.
Umsebenzi ohlangene ngenxa ye-LiteRT
Ukuze kuvinjelwe onjiniyela ukuthi bangaphambani ngokulungiselela okuyinkimbinkimbi, i-Google ihlanganise i-Tensor SDK ne I-LiteRTLolu hlaka lusebenza njengesendlalelo sokucashunwa esisusa isidingo sokubhekana nama-SDK aqondene nomthengisi noma abahlanganisi abayinkimbinkimbi, okunikeza i-API elula nehambisanayo yokufaka amamodeli okufunda komshini emaphethelweni.
Le nqubo ihlukaniswe ngezigaba ezintathu ezibalulekile. Okokuqala kukhona ukuhlanganiswa kwamamodelilapho ungaguqula khona amaphrojekthi akho e-PyTorch noma I-TFLite isekelwe ku-TensorFlow kuma-binary alungiselelwe kusetshenziswa i-LiteRT Torch. Bese kufika i- ukuthunyelwa, kusetshenziswa i-Play Feature Delivery kanye nalokho okubizwa ngokuthi ama-AI Packs ukuze kusatshalaliswe kahle imitapo yolwazi kanye namamodeli ahlanganisiwe ngaphakathi kohlelo lokusebenza.
Ekugcineni, sifika ekuqalisweni kwesiphetho. Ngenxa Isikhathi Sokusebenza se-LiteRTUngenza imodeli yakho isebenze ku-TPU ngemigqa embalwa yekhodi. Okuhle kunakho konke, uhlelo luhlakaniphile: uma ngesizathu esithile i-TPU ingatholakali, ungalulungiselela izindlela zokusekela ukuze umthwalo udluliselwe ngokuzenzakalelayo ku-CPU noma ku-GPU, okuqinisekisa ukuthi uhlelo lokusebenza alulokothi luqine.
Ingadi Yemodeli: Ikhathalogi yamathuba
Asikho isidingo sokuqala kusukela ekuqaleni, njengoba i-Beta SDK ifaka phakathi Imodeli Yengadi Kuyamangalisa. Liyilabhulali enamamodeli angaphezu kwe-100, kokubili i-ML yakudala kanye ne-AI ekhiqizayo, kufaka phakathi izinguqulo ze I-Gem 3 1BNgaphezu kwenani elikhulu lamamodeli ahlanganiswe kusengaphambili ongawalanda ngqo emphakathini we-Hugging Face ku-LiteRT.
Uma ufuna ukudala imisebenzi yombhalo, amamodeli amancane olimi afana Umsebenzi we-Gemma Zikuvumela ukuthi wenze izenzo zasendaweni ngaphakathi kohlelo lokusebenza, kuyilapho i-EmbeddingGemma ingeza amakhono athuthukile okuqonda. Ngakolunye uhlangothi, i-SDK ikuvumela ukuthi usebenzise ukutholwa kwezinto kanye nokumapha okujulileokuyigolide elicwengekile lezinhlelo zokusebenza zekhamera ezidinga ukusabela endaweni yomsebenzisi ngesikhathi sangempela.
Abakakhohlwa ngomsindo kanye nokufinyeleleka. Manje sekungenzeka ukusebenzisa ukuqashelwa kwezwi kusukela ekuqaleni kuya ekugcineniLokhu kuqinisekisa ukulotshwa kwamagama ngamathuluzi okubambezeleka aphansi kakhulu kanye nokuhumusha asebenza ungaxhunyiwe ku-inthanethi, okugcina ubumfihlo bedatha njengoba kungekho lutho oluphuma efonini.
Ukuthuthukiswa kobuchwepheshe kanye nokusekelwa kwehadiwe
Ukuze uthole okuningi kukho, kubalulekile ukwazi ukuthi iyiphi ihadiwe esekelwayo. Njengamanje, uhlelo lwe-ecosystem lugxile emndenini. I-Pixel 10kufaka phakathi amamodeli e-Pro, Pro XL kanye ne-Pro Fold, wonke afakwe i-SoC I-Google Tensor G5Ukuze i-AI isebenze kahle, i-Google incoma ukusebenzisa amafulegi athile okuthuthukisa ngesikhathi sokuhlanganiswa.
Isibonelo, ekugelezeni kwe-LiteRT Python, kuvamile kakhulu ukusebenzisa ifulegi i-google_tensor_truncation_type=”isigamu” ukulungisa ukusebenza kanye nokusetshenziswa kwezinsiza. Endabeni yamamodeli olimi abanzi (i-LLM), ukuthumela ngaphandle kudinga amapharamitha anemininingwane, njengokucushwa kwe iresiphi_ye-quantization kanye nokusebenza kokusekelwa kwamamodeli amakhulu ngesichazamazwi sokucushwa kwe-AOT.
Kubalulekile ukusho ukuthi, nakuba i-NNAPI yayikhona ngaphambilini, isiphelelwe yisikhathi kusukela ngo-1998. I-Android 15Isu manje ukudlulisa konke kubathunywa be-LiteRT, lapho ukusekelwa kwe-Pixel TPU sekuyinto ebalulekile ekushintsheni ukuqaliswa okudala nokuthola indawo. ukusebenza kahle kwamandla ngaphezulu.
Uguquko lwe-Gemma 4 kanye ne-Multimodal AI
Ake sixoxe ngezinto zakamuva: ukufika kwe I-Gem 4 12BNgokungafani namanye amamodeli anamathisela isifaki-khodi sesithombe kumodeli yolimi, i-Gemma 4 icubungula izithombe ngokwemvelo. Lo mklamo olula awugcini nje ngokunciphisa ukusetshenziswa kwe-VRAM kodwa futhi uvumela... ukucabanga okuphambene nemodi okuguquguqukayo kakhulu futhi okunobungane.
Ngefasitela lomongo le- 256K amathokheniLo modeli ungaphatha izingxoxo ezinde ngezithombe eziningi ngaphandle kokulahlekelwa umkhondo. Ngaphezu kwalokho, usatshalaliswa ngaphansi kwelayisensi I-Apache 2.0Iguquguquka kakhulu ekusetshenzisweni kwezentengiselwano nasekusatshalalisweni kabusha, ivumela i-AI yezindlela eziningi ukuthi isebenze kuma-laptops esimanje kusetshenziswa ama-quantizations angu-4 noma angu-8 bit.
Umgomo we-Google ucacile: ufuna abathuthukisi basebenzise amamodeli ayo okulinganisa isisindo ukuze balawule uhlelo lwendalo. Ngokwenza kube lula ukuthi i-AI ibe wendawo futhi enamandlaZinciphisa ukuthembela kuma-API angaphandle futhi zakha umphakathi ohlangene ozungeze ihadiwe ye-Tensor kanye nesofthiwe ye-LiteRT.
Uhlelo lokuthuthukiswa kwe-Pixel luthathe igxathu elibalulekile ngokuhlanganisa ihadiwe ye- I-Tensor G5 Ngokusebenziseka kwe-LiteRT kanye namandla amamodeli afana ne-Gemma 4, ngenxa yokushintshela ku-Beta ye-SDK kanye nokutholakala kwekhathalogi enkulu yamamodeli alungiselelwe kusengaphambili, ukudala izinhlelo zokusebenza ezicubungula umbono, inkulumo, kanye nolimi ngasese nangokushesha kakhulu manje sekuyinto efinyelelekayo kunoma yimuphi umhleli wezinhlelo.
