Installing Tensorflow in WSL2

tensorflow_wsl2.md

Using conda to install cuda dependencies:

conda install -c conda-forge cudatoolkit=11.8.0
conda install -c "nvidia/label/cuda-11.8.0" cuda-nvcc

Install tensorflow and cuda utilities:

python3 -m pip install nvidia-cudnn-cu11==8.6.0.163 tensorflow==2.13.*

Copy libdevice library:

mkdir -p $CONDA_PREFIX/lib/nvvm/libdevice
cp $CONDA_PREFIX/pkgs/cudatoolkit-11.8.0-<tag>/lib/libdevice.10.bc $CONDA_PREFIX/lib/nvvm/libdevice/

Edit environment variables:

echo 'CUDNN_PATH=$(dirname $(python -c "import nvidia.cudnn;print(nvidia.cudnn.__file__)"))' >> $CONDA_PREFIX/etc/conda/activate.d/env_vars.sh
echo 'export LD_LIBRARY_PATH=/usr/lib/wsl/lib/:$CUDNN_PATH/lib:$CONDA_PREFIX/lib/:$LD_LIBRARY_PATH' >> $CONDA_PREFIX/etc/conda/activate.d/env_vars.sh
echo 'export XLA_FLAGS=--xla_gpu_cuda_data_dir=$CONDA_PREFIX/lib' >> $CONDA_PREFIX/etc/conda/activate.d/env_vars.sh

Verify installation:

python3 -c "import tensorflow as tf; print(tf.config.list_physical_devices('GPU'))"