PyTorch, a popular machine learning framework, has been updated to version 1.5.0. This release includes several major additions and improvements to the API, including:
-
The C++ API, previously considered experimental, has finally been stabilized. Users can now easily migrate their models from the Python API to the C++ API.
-
The torch.distributed.rpc package has been stabilized, providing a wide range of features in distributed learning, including automatic calculation of gradients and updating model parameters.
-
Updated torch_xla, a package that uses the XLA compiler to speed up model training on cloud TPUs.
-
The torchaudio, torchvision and torchtext packages have also been updated, providing tools for developing models that process sound, graphic and text data.
-
Python 2 is no longer supported. All future development will be for Python 3 only.
Source: linux.org.ru