Skip to content

Latest commit

 

History

History
40 lines (30 loc) · 1.08 KB

install_tf28.md

File metadata and controls

40 lines (30 loc) · 1.08 KB
Raw

Install TensorFlow 2.8 (or 2.9) + Pillow + Tqdm guide

For training, we recommend using the NVIDIA GPU with architecture ≥ volta (e.g. v100, any 20 or 30 series).

For NVIDIA GPU drvier, we recommend using the lastest version. CUDA installation is not required for conda user (See below).

Installation step by step

For conda user (recommend)

  1. Create new conda environment
conda create -n tf28 python=3.8
  1. Activated the new environment
conda activate tf28
  1. Install cudatoolkit 11.2, cudnn 8.1, tqdm and pillow
conda install cudatoolkit=11.2 -c conda-forge
conda install cudnn=8.1 -c conda-forge
conda install tqdm
conda install pillow
  1. Install TensorFlow 2.8
pip install tensorflow==2.8.0
  1. (optional) In some cases, you may have to export LD_LIBRARY_PATH to help TensorFlow find CUDA, etc.
export LD_LIBRARY_PATH=$CONDA_PREFIX/lib

You may also add this script into the conda script, see https://conda.io/projects/conda/en/latest/user-guide/tasks/manage-environments.html#saving-environment-variables

If 5 is still not working, contact me :)