MLPerf Mobile

MLCommons

Mayelana nalolu hlelo lokusebenza

I-MLPerf Mobile iyithuluzi lokulinganisa lomthombo ovulekile lamahhala elakhelwe ukukala ukusebenza kwamadivayisi eselula njengama-smartphone namathebulethi emisebenzini ehlukahlukene yobuhlakani bokwenziwa (AI) nokufunda komshini (ML). Imithwalo yomsebenzi ehloliwe ihlanganisa ukuhlukaniswa kwezithombe, ukuqonda ulimi, ukukhuphuka kokucaca okuphezulu, nokukhiqizwa kwezithombe ngokusekelwe ekwazisweni kombhalo. Le bhentshimakhi isebenzisa ukusheshisa kwezingxenyekazi zekhompiyutha ze-AI kumadivayisi eselula amaningi akamuva ukuze kuqinisekiswe ukusebenza okuhle kakhulu lapho kungenzeka khona.

I-MLPerf Mobile yakhiwe futhi inakekelwa iqembu elisebenzayo le-MLPerf Mobile kwa-MLCommons®, inhlangano engenzi nzuzo ye-AI/ML yobunjiniyela eyakhiwe ngamalungu angu-125+ okuhlanganisa amafemu emboni nezifundiswa ezivela inkimbi yezikhungo ezihlukene emhlabeni jikelele. I-MLCommons ikhiqiza amabhentshimakhi asezingeni lomhlaba okuqeqeshwa kwe-AI kanye nencazelo kuzo zonke izikali zesistimu, kusukela ekufakweni okukhulu kwesikhungo sedatha kuya kumadivayisi amancane ashumekiwe.

Izici ze-MLPerf Mobile zifaka:

- Ukuhlolwa kwe-Benchmark kuzo zonke izizinda ezihlukahlukene ngokusekelwe kumamodeli e-AI asezingeni eliphezulu, okuhlanganisa:

- Ukuhlukaniswa kwezithombe
- Ukutholwa kwento
- Ukuhlukaniswa kwesithombe
- Ukuqonda ulimi
- Super resolution
- Ukukhiqizwa kwesithombe kusuka emiyalweni yombhalo

- Ukusheshisa okwenziwe ngokwezifiso kwe-AI kumadivayisi eselula akamuva nama-SoCs.

- Ukusekelwa okubanzi kwamadivayisi we-Android nge-TensorFlow Lite ehambisa ukusheshiswa kokubuyela emuva.

- Izindlela zokuhlola ezenzelwe wonke umuntu kusukela kubasebenzisi abavamile abafuna ukuhlolwa kokusebenza okusheshayo kumalungu e-MLCommons ahlose ukuhambisa imiphumela esemthethweni ukuze ishicilelwe.

- Ukubambezeleka okupholisa phansi okwenziwa ngendlela oyifisayo phakathi kokuhlolwa ukuze kugwenywe ukunyakaziswa okushisayo nokuqinisekisa imiphumela enembile.

- Isitoreji semiphumela esisekelwe emafini ongasikhetha ukuze ukwazi ukulondoloza futhi ufinyelele imiphumela yakho yangaphambilini kumadivayisi amaningi endaweni eyodwa. (Lesi sici simahhala kodwa sidinga ukubhaliswa kwe-akhawunti.)

I-MLPerf Mobile ivamise ukubuyekezwa izikhathi eziningi ngonyaka ngokuhlolwa okusha nokusekelwa kokusheshisa njengoba amamodeli e-AI namandla ezingxenyekazi zekhompiyutha eguqukayo. Sicela uqaphele ukuthi okunye ukuhlola kwebhentshimakhi kungase kungasekelwe, ngakho-ke kungase kungabonisi njengokutholakalayo ukuze kuhlolwe, kumadivayisi amadala.

Ikhodi yomthombo kanye nemibhalo yohlelo lokusebenza lwe-MLPerf Mobile kuyatholakala ku-MLCommons Github repo. Ukuze uthole ukwesekwa komsebenzisi noma imibuzo, sicela ukhululeke ukuvula izinkinga ku-Github repo yohlelo lokusebenza:

github.com/mlcommons/mobile_app_open

Uma wena noma inhlangano yakho nithanda ukuba yilungu le-MLCommons, sicela nithinte [e-mail] ukuze uthole ulwazi olwengeziwe.

Yini entsha

Added support for Exynos 2500
For support, please open an issue in the MLPerf Mobile GitHub repo.