Release notes

For information on available Radeon Software for Linux releases, please refer to Linux® Drivers for AMD Radeon™ and Radeon PRO™ Graphics.

ROCm PyTorch wheel install instructions

#For PyTorch 2.1.2 + torchvision 0.16.1
pip3 install torch==2.1.2 torchvision==0.16.1 -f https://repo.radeon.com/rocm/manylinux/rocm-rel-6.1/

#For PyTorch 2.0.1 + torchvision 0.15.2
pip3 install torch==2.0.1 torchvision==0.15.2 -f https://repo.radeon.com/rocm/manylinux/rocm-rel-6.1/

#For PyTorch 1.13.1 + torchvision 0.14.0
pip3 install torch==1.13.1 torchvision==0.14.0 -f https://repo.radeon.com/rocm/manylinux/rocm-rel-6.1/

#For PyTorch 1.10.1 + torchvision 0.11.2
pip3 install torch==1.10.1 torchvision==0.11.2 -f https://repo.radeon.com/rocm/manylinux/rocm-rel-6.1/

-------------------------------------------------------------------------------------------
Note: A compatible pytorch-triton-rocm package will also be installed as a dependency when the version of PyTorch wheels is greater than 2.0.

Using MIOpen kdb files with ROCm PyTorch wheels

MIOpen .kdb files contain precompiled kernels that can speed up the warm-up phase of an application. More information here: installing-miopen-kernels-package

Starting from ROCm5.6, MIOpen .kdb files can be used with ROCm PyTorch wheels. However, the .kdb files need to be placed in a specific location wrt the PyTorch installation path. We provide a helper script that simplifies this task for the user. The script takes in the ROCm version and user's GPU architecture as inputs, and works for Ubuntu and CentOS.

NOTE: This is an "alpha" feature.

Helper script: install_kdb_files_for_pytorch_wheels.sh

Usage:
After installing ROCm PyTorch wheels:

1. [Optional] export GFX_ARCH=gfx90a
2. [Optional] export ROCM_VERSION=6.1
3. ./install_kdb_files_for_pytorch_wheels.sh