PyTorch Release 18.08
The NVIDIA container image of PyTorch, release 18.08, is available.
Contents of PyTorch
This container image contains the complete source of the version of PyTorch in
/opt/pytorch. It is pre-built and installed in the
pytorch-py3.6 Conda™ environment in the container image.
The container also includes the following:
- Ubuntu 16.04 including Python 3.6 environment
- NVIDIA CUDA 9.0.176 (see Errata section and 2.1) including CUDA® Basic Linear Algebra Subroutines library™ (cuBLAS) 9.0.425
- NVIDIA CUDA® Deep Neural Network library™ (cuDNN) 7.2.1
- NCCL 2.2.13 (optimized for NVLink™ )
- Caffe2 0.8.1
- DALI 0.1.2 Beta
- Tensor Core Optimized Examples:
Release 18.08 is based on CUDA 9, which requires NVIDIA Driver release 384.xx.
Key Features and Enhancements
- PyTorch container image version 18.08 is based on PyTorch 0.4.1. PyTorch 0.4.1 is released and included with this container. See the release notes at https://github.com/pytorch/pytorch/releases for significant changes from PyTorch 0.4.
- Apex is now entirely Python for improved compatibility. Previous versions of Apex will not work with PyTorch 0.4.1 or newer versions.
- New ops in 18.08:
- Support for cross-device gradient clipping.
torch.eigin CUDA have been fixed, previously they could return incorrect results.
- An implementation of GNMT v2. The GNMT v2 model is similar to the one discussed in the Google's Neural Machine Translation System: Bridging the Gap between Human and Machine Translation paper.
- Latest version of cuDNN 7.2.1.
- Latest version of DALI 0.1.2 Beta.
- Added support for the Tensor Core Optimized Example: PyTorch GNMT model
- Ubuntu 16.04 with July 2018 updates
PYTHONPATH environment variable in this container version has been updated to include all packages installed in the Conda environment and all PyTorch related packages. Users that rely on
PYTHONPATH to point to local modules are advised to carefully check and set their
PYTHONPATH variable in this container and moving forward.
Tensor Core Examples
An implementation of GNMT v2. The GNMT v2 model is similar to the one discussed in the Google's Neural Machine Translation System: Bridging the Gap between Human and Machine Translation paper.
The DALI integrated ResNet-50 samples in the 18.08 NGC TensorFlow and PyTorch containers may result in lower than expected performance results. We are working to address the issue in the next release.