PyTorch released version 1.2.0, and 1.1, the new version is mainly to add / change features include:
- TorchScript perfect environment provides a new model will compile the API TorchScrip
- Extended support the export of model ONNX
- Enhanced support for Transformer module-level architecture
- In addition, TensorBoard now can quickly start using the
PyTorch is a Torch7 team open source Python priority of deep learning framework that provides two advanced features:
-
Tensor powerful GPU accelerated computing (similar numpy)
-
Construction of Neural Network Based on the depth of tape automated system upgrades
You can reuse your favorite python package, such as numpy, scipy and Cython, expand PyTorch when needed.