Kwenzekeni
I-AWS ishicilele umhlahlandlela wobuchwepheshe futhi yakhulula Isitsha Sokufunda Okujulile (i-DLC) se-WhisperX, isandiso esinomthombo ovulekile semodeli ye-Whisper ye-OpenAI. Lesi sitsha siklanyelwe ukuthunyelwa ku-Amazon SageMaker AI, esekela kokubili isikhathi sangempela namaphoyinti okugcina asynchronous. Uhlelo lwengeza ukudayela kwesipikha kanye nezitembu zesikhathi zeleveli yegama ekulotshweni okujwayelekile, kubhekwana nemikhawulo kumathuluzi ajwayelekile enkulumo-kuya-kumbhalo wemisebenzi yebhizinisi.
I-AWS yethule i- Container (DLC) ye-WhisperX, etholakalayo ukuze isetshenziswe ku-Amazon SageMaker AI. I-WhisperX iwumthombo ovulekile wephrojekthi enweba imodeli ye-OpenAI ye-Whisper yokuqaphela inkulumo ngokuzenzakalelayo ngokungeza ukucatshangelwa kweqoqo, ukuqondanisa okuphoqelelwe kwe-wav2vec2 kwezitembu zesikhathi zegama ngalinye, kanye nokudayela kwesipikha. I-AWS DLC iphasela lezi zingxenye esithombeni esilungele i-GPU esilandela inkontileka yokusebenza ye-SageMaker AI evamile, esusa isidingo sabasebenzisi sokwakha izithombe zangokwezifiso noma ukuphatha amathokheni e-Hugging Face.
Ukuthunyelwa kusekela izinhlobo ezimbili zephoyinti lokugcina: isikhathi sangempela kanye ne-asynchronous. Iphoyinti lokugcina lesikhathi sangempela lifanele iziqeshana ezimfushane, ezisebenzisanayo eziqeda ngaphakathi kwekhephu yokuphendula yemizuzwana engama-60 ye-SageMaker AI, ebuyisela okulotshiweyo ngokuvumelana. Indawo yokugcina engavumelanisiwe inconywa kumafayela omsindo amade, njengoba ithengisa okokufaka nokukhiphayo nge-Amazon S3, isusa umkhawulo wesikhathi futhi ivumela ukucubungula okuyinkimbinkimbi. Zombili izindawo zokugcina zidinga ukucushwa okuqondile, okuhlanganisa ukuphina i-InferenceAmiVersion ku-al2-ami-sagemaker--gpu-3-1 ukuze kuvinjelwe amaphutha okuqalisa.
Umthombo uhlinzeka ngebhukwana le-JupyterLab elisebenzisekayo endaweni yokugcina yamasampula ye-AWS ukuze kuhlolwe ukuthunyelwa. Isibonelo sisebenzisa ukurekhodwa kwesizinda somphakathi sokuxhumana kokulawulwa kwethrafikhi yendiza kusuka ku-US Airways Flight 1549 ukuze kuboniswe ikhono lesistimu lokusingatha ukushintshaniswa kwerediyo yezinkampani eziningi ngomsindo wangemuva. Okukhiphayo kufaka phakathi amasegimenti, izitembu zesikhathi zegama ngalinye, namalebula esipikha, abalulekile ezinhlelweni zokusebenza ezidinga ukuhlaziywa komsindo okunembayo kanye nokuhlolwa kokuthobela imithetho.
Imininingwane yomthombo: aws.amazon.com ↗
Kungani kubalulekile
Lokhu kukhishwa kwehlisa umgoqo wokuthi amabhizinisi asebenzise ukuhlaziywa komsindo othembeke kakhulu ngokupakisha izingxenye eziyinkimbinkimbi ezifana nokuqondanisa kwe-wav2vec2 kanye nokudayela kusithombe esiphethwe, esilungele i-GPU. Ivumela amaqembu ukuthi asebenzise okulotshiweyo okunelebula yesipika ngaphandle kokwakha iziqukathi zangokwezifiso, ukusekela izimo zokusetshenziswa ezikhungweni zokuxhumana, ukutholwa kwezomthetho, namagama-ncazo emidiya. Ngokuhlinzeka ngenkontileka yokusebenzela esezingeni, i-AWS ihlanganisa amakhono athuthukile omthombo ovulekile we-AI nengqalasizinda yayo yamafu ephethwe, yenza kube lula ukucutshungulwa komsindo okunokwehla nokuthobelayo.
Amathuluzi ajwayelekile enkulumo-kuya-umbhalo ngokuvamile ehluleka ukunikeza ukuhlonza okuthembekile kwesipika kanye nezitembu zesikhathi ezinembile, ezibalulekile ezinhlelweni zebhizinisi ezifana nokuhlaziywa kwesikhungo sokuxhumana, ukutholwa kwezomthetho, namagama-ncazo emidiya. Ngokupakisha i-WhisperX ku-DLC ephethwe, i-AWS inikeza isisombululo esilungele ukukhiqiza esibhekana nalezi zikhala ngaphandle kokudinga ubuchwepheshe obujulile bengqalasizinda.
Ukuhlanganiswa kokudayela kwesipikha kanye nokuqondanisa kweleveli yegama kunika amandla ukuhlaziya okuyimbudumbudu okwengeziwe kwedatha yomsindo. Isibonelo, izikhungo zokuxhumana zingakala isikhathi sokukhuluma nokubambelela kweskripthi, kuyilapho amaqembu emidiya engakwazi ukukhiqiza amafayela anembile e-SubRip Subtitle (SRT) kanye namathrekhi ombhalo wevidiyo yewebhu (VTT). Leli khono lisekela ukuthotshelwa kwemithetho emikhakheni yezokunakekelwa kwempilo, yezomthetho, neyezezimali lapho kudingeka khona ukuchazwa kwenkulumo okunembile ukuze kucwaningwe.
Ukukhishwa kugqamisa inkambiso yabahlinzeki bamafu abahlanganisa amamodeli akhethekile wemithombo evulekile ye-AI kumasevisi abo aphethwe. Ngokunikeza isiqukathi esimisiwe, i-AWS yehlisa ukusebenza okungaphezulu ukuze kusetshenziswe amamodeli okuqaphela inkulumo athuthukile, okuwenza afinyeleleke kakhudlwana kuhlu olubanzi lonjiniyela namabhizinisi.
I-Interactive Mechanism: Indlela Esebenza Ngayo Ngempela
Hlola ubuchwepheshe obuyisisekelo ngemuva kwalokhu kuthuthukiswa ngokuhlanganyela.
In AI, what are a model's "parameters"?
Ongakubuka ngokulandelayo
Gada amamethrikhi okwamukelwa kwekhasimende kanye namamodeli entengo athile ezimweni ze-GPU ezidingekayo ukuze kusetshenziswe lesi siqukathi. Buka izibuyekezo zephrojekthi yomthombo ovulekile we-WhisperX ezingase ziboniswe ezinguqulweni zesiqukathi ezizayo, futhi ubheke ukuthi le phethini yokuthunyelwa iyithonya kanjani imakethe ebanzi yamasevisi aphethwe enkulumo-to-umbhalo.
Umthombo awucacisi intengo yezimo ze-GPU (ml.g4dn.xlarge noma ml.g5.2xlarge) edingekayo ukuze kuqalise isiqukathi, futhi awunikezi imininingwane yezindleko zokukhokhiswa okuqhubekayo kwamaphoyinti wokugcina we-GPU. Abasebenzisi kufanele baqaphe AWS izibuyekezo zentengo nezici zokulawula izindleko ze-SageMaker AI.
Izibuyekezo zesikhathi esizayo zephrojekthi yomthombo ovulekile we-WhisperX zingase zethule izici ezintsha noma ukuthuthukiswa okuboniswa ezinguqulweni ezilandelayo ze-AWS DLC. Ukulandelela lezi zibuyekezo kuzobaluleka kubasebenzisi abathembele emandleni akamuva ekuhlanganiseni izipika nokuqondanisa.
Ukwamukelwa kwalesi sitsha ngamakhasimende ebhizinisi kuzobonisa isidingo semakethe samasevisi aphethwe, athembekile kakhulu enkulumo-kuya-umbhalo. Ukubona ukuthi i-AWS ilimisa kanjani leli thuluzi ngokumelene nokunye okunikezwayo kokuqaphela inkulumo kuzohlinzeka ngemininingwane ngendawo yokuncintisana.