O le PyTorch, le ta'uta'ua fa'aa'oa'oga fa'aa'oa'oga masini, ua fa'afouina i le version 1.3.0 ma o lo'o fa'aauau pea ona maua le malosi ma lona taula'i i le tautuaina o mana'oga o tagata su'esu'e ma tagata fai polokalame.
O nisi suiga:
- lagolago fa'ata'ita'i mo fa'aigoa tensor. E mafai nei ona e va'ai ile tensor dimensions ile igoa, nai lo le fa'ailoaina o se tulaga atoatoa:
NCHW = [‘N’, ‘C’, ‘H’, ‘W’] images = torch.randn(32, 3, 56, 56, names=NCHW)
images.sum('C')
images.select('C', index=0) - lagolago mo le 8-bit quantization faʻaaogaina FBGEMM и QNNPACK, o loʻo tuʻufaʻatasia i le PyTorch ma faʻaoga se API masani;
- galue mo masini feaveaʻi faʻatautaia iOS ma Android;
- fa'amatu'u atu o mea faigaluega fa'aopoopo ma faletusi mo fa'amatalaga fa'ata'ita'iga.
I le faaopoopo atu, lomia pueina o lipoti mai le taimi ua mavae Pytorch Developer Conference 2019.
puna: linux.org.ru