pip install --extra-index-url https://developer.download.nvidia.com/compute/redist --upgrade nvidia-dali-cuda110==1.26.0
pip install torch==1.8.2 torchvision==0.9.2 torchaudio==0.8.2 --extra-index-url https://download.pytorch.org/whl/lts/1.8/cu111
pip install vit-pytorch==0.26.7
pip install git+https://github.com/wbaek/theconf.git[https://github.com/omnia-unist/FusionFlow_PyTorch]
git clone --recursive https://github.com/pytorch/pytorch
cd pytorch
# if you are updating an existing checkout
git submodule sync
git submodule update --init --recursive
pip3 install torch==1.8.2 torchvision==0.9.2 torchaudio==0.8.2 --extra-index-url https://download.pytorch.org/whl/lts/1.8/cu111
(install vit-pytorch kind of source build)
pip unistall torch (clean only torch)
python setup.py develop
[https://github.com/omnia-unist/FusionFlow_DALI]
cd wheelhouse
pip install nvidia_dali_cuda110-1.26.0.dev0-12345-py3-none-manylinux2014_x86_64.shl
If you want to install from the source code
- When compliation, using CUDA 11.0, CUDA 11.8 and CUDA 12.1 (other 11.x is not supported)
Question about installation on DALI with dockerfile · Issue #4814 · NVIDIA/DALI
- Compilation at the docker directory
sudo CUDA_VERSION=11.8 PYVER=3.8 BUILD_TEST=0 ./build.sh
- Made .wheel file to install (uninstall is necessary when the package is already installed
pip uninstall nvidia-dali-cuda110pip install ../wheelhouse/nvidia_dali_cuda110-1.26.0.dev0-12345-py3-none-manylinux2014_x86_64.whl- ./train.sh
- Execute Main Experiment
$ ./train.sh