05/02/2024
- Added PyTorch 2.3.0 Support: We now added PyTorch 2.3.0 with CUDA 11.8 - 12.1, Python 3.8 - 3.11 variance.
03/06/2024
- Added PyTorch 2.2.0 Support: We now added PyTorch 2.2.0 with CUDA 11.8 - 12.1, Python 3.8 - 3.10 variance.
12/15/2023
- Added AWS-OFI-NCCL 1.7.4: We now added AWS-OFI-NCCL 1.7.4 release as a dependency of PyTorch 2.1.0 conda package. No extra configuration needed for PyTorch to use EFA on P5 instances. To upgrade aws-ofi-nccl without upgrading PyTorch binaries, use this command:
mamba install aws-ofi-nccl -c https://aws-ml-conda.s3.us-west-2.amazonaws.com
.
10/18/2023
- Added PyTorch 2.1.0 Support: We now added PyTorch 2.1.0 with NCCL 2.18.5 and CUDA 11.8 - 12.1, Python 3.8 - 3.10 variance.
- Added AWS-OFI-NCCL 1.7.3 Support: We now added AWS-OFI-NCCL 1.7.3 as dependency of PyTorch 2.1.0 Conda packages. No extra configuration needed for PyTorch to use EFA.
09/27/2023
- Added CUDA 12.2, CUDNN 8.9.4 and NCCL 2.18.5 PyTorch build, this update brings best performance on P5 instances.
09/02/2023
- Added AWS-OFI-NCCL 1.7.2: We now added AWS-OFI-NCCL 1.7.2 release as a dependency of PyTorch 2.0.1 conda package. This release contains fixes for P5/H100 systems. No extra configuration needed for PyTorch to use EFA on P5 instances.
08/18/2023
- PyTorch support for AWS P5 instances: added PyTorch 2.0.1 built with CUDA 12.1 and NCCL 2.18.3 that works best on P5 instances.
- Added AWS-OFI-NCCL 1.7.1: We now added AWS-OFI-NCCL 1.7.1 release as a dependency of PyTorch 2.0.1 conda package. No extra configuration needed for PyTorch to use EFA on P5 instances.
06/05/2023
- Added PyTorch 2.0.1 Support: We now added PyTorch 2.0.1 with CUDA 11.7 - 11.8, Python 3.8 - 3.10 variance.
- Added AWS-OFI-NCCL 1.6.0 Support: We now added AWS-OFI-NCCL 1.6.0 as dependency of PyTorch 2.0.1 conda packages.
03/30/2023
- Added PyTorch 2.0.0 Support: We now added PyTorch 2.0.0 with CUDA 11.7 - 11.8, PyThon 3.8 - 3.10 variance.
03/02/2023
- Added PyTorch 1.13.1 Support: We now added PyTorch 1.13.1 with CUDA 11.6 - 11.7, PyThon 3.7 - 3.10 variance.
- Added AWS-OFI-NCCL 1.5.0 Support: We now added AWS-OFI-NCCL 1.5.0 as dependency of PyTorch 2.0.0 conda packages. This new release no longer requires a fixed NCCL version as dependency. For more details see AWS OFI NCCL Plugin documentation.
- Updated User Guide on faster dependency resolving time: We updated installation guide to use
-c nvidia/label/cuda-$CUDA_VERSION
for faster conda dependency resolver time.
02/07/2023
- Expanded CUDA version Support: We now added CUDA 11.5 support for pytorch-1.12.1, and CUDA 11.6 support for pytorch-1.11.0.
11/29/2022
- Expanded PyTorch Version Support: We now added pytorch-1.13.0 support in addition to the existing pytorch-1.11.0 and pytorch-1.12.1 support.
- Expanded CUDA version Support: We now added CUDA 11.7 support for pytorch-1.13.0, pytorch-1.13.0 comes with both CUDA 11.6 and CUDA 11.7
- Expanded Python Version Support: pytorch-1.11.0, pytorch-1.12.1, and pytorch-1.13.0 all include builds for python 3.7, 3.8, 3.9, and 3.10.
11/07/2022
- Expanded PyTorch Version Support: We now added pytorch-1.11.0 support in addition to the existing pytorch-1.12.1 support.
- Expanded Python Version Support: Both pytorch-1.11.0 and pytorch-1.12.1 include builds for python 3.7, 3.8, 3.9, and 3.10.
08/24/2022
- Initial Release: Added support for pytorch-1.12.1 for cuda-11.6 and python-3.8.