PyTorch, the popular open-source machine learning framework, has been updated to version 1.3.0 and continues to gain momentum with its focus on meeting the needs of both researchers and application programmers.
Some changes:
- experimental support for named tensors. Now you can refer to tensor dimensions by name, instead of specifying an absolute position:
NCHW = ['N', 'C', 'H', 'W'] images = torch.randn(32, 3, 56, 56, names=NCHW)
images.sum('C')
images.select('C', index=0) - support for 8-bit quantization with FBGEMM ΠΈ QNNPACK, which are integrated into PyTorch and use a common API;
- work on mobile devices running iOS and Android;
- release of additional tools and libraries for interpreting models.
Additionally, published recording of reports from the last conference Pytorch Developer Conference 2019.
Source: linux.org.ru