Fixes for GLIBC errors when installing tensorflow or pytorch on older Red Hat or CentOS cluster environments

# Conda Installation First create the following ``` # >>> conda initialize >>> # !! Contents within this block are managed by 'conda init' !! __conda_setup="$('conda' 'shell.bash' 'hook' 2> /dev/null)" if [ $? -eq 0 ]; then eval "$__conda_setup" else if [ -f "$EBROOTANACONDA3/etc/profile.d/conda.sh" ]; then . "$EBROOTANACONDA3/etc/profile.d/conda.sh" else export PATH="$EBROOTANACONDA3/bin:$PATH" fi fi unset __conda_setup # <<< conda initialize <<< ``` and save this script`conda.sh` in `$HOME` directory. To install conda python package (Tensorflow for example) login to the compute node. First logon to Lochness and then `srun --partition=datasci --ntasks=1 --mem=10G --pty bash` ``` node430-41 ~ >: module load Anaconda3 node430-41 ~ >: conda create --name tf python=3.9 Collecting package metadata (current_repodata.json): done Solving environment: done ## Package Plan ## environment location: /home/g/guest24/miniconda3/envs/tf added / updated specs: - python=3.9 The following packages will be downloaded: <output snipped> Proceed ([y]/n)?y <output snipped> # # To activate this environment, use # # $ conda activate tf # # To deactivate an active environment, use # # $ conda deactivate ``` Activate the new 'tf' environment ``` node430-41 ~ >: source $HOME/conda.sh node430-41 ~ >: conda activate tf (tf) node430-41 ~ >: ``` Install tensorflow-gpu ``` (tf) node430-41 ~ >: conda install -c anaconda tensorflow-gpu Collecting package metadata (current_repodata.json): done Solving environment: done ## Package Plan ## environment location: /home/g/guest24/miniconda3/envs/tf added / updated specs: - tensorflow-gpu <output snipped> The following packages will be SUPERSEDED by a higher-priority channel: ca-certificates pkgs/main --> anaconda certifi pkgs/main --> anaconda openssl pkgs/main --> anaconda Proceed ([y]/n)?y <output snipped> mkl_fft-1.1.0 | 143 KB | ####################################################################################### | 100% urllib3-1.25.9 | 98 KB | ####################################################################################### | 100% cudatoolkit-10.1.243 | 513.2 MB | ####################################################################################### | 100% protobuf-3.12.3 | 711 KB | ####################################################################################### | 100% blinker-1.4 | 21 KB | ####################################################################################### | 100% requests-2.24.0 | 54 KB | ####################################################################################### | 100% werkzeug-1.0.1 | 243 KB | ####################################################################################### | 100% Preparing transaction: done Verifying transaction: done Executing transaction: done ``` Check to see if tensorflow can be loaded ``` (tf) node430-41 ~ >: python Python 3.9.13 (main, Oct 13 2022, 21:15:33) [GCC 11.2.0] :: Anaconda, Inc. on linux Type "help", "copyright", "credits" or "license" for more information. >>> ``` Simple tensorflow test program to make sure the virtual env can access a gpu. Program is called `tf.gpu.test.py` ``` import tensorflow as tf if tf.test.gpu_device_name(): print('Default GPU Device: {}'.format(tf.test.gpu_device_name())) else: print("Please install GPU version of TF") ``` Slurm script to submit the job ``` #!/bin/bash -l #SBATCH --job-name=tf_test #SBATCH --output=%x.%j.out # %x.%j expands to JobName.JobID #SBATCH --nodes=1 #SBATCH --tasks-per-node=1 #SBATCH --partition=datasci #SBATCH --gres=gpu:1 #SBATCH --mem=4G # Purge any module loaded by default module purge > /dev/null 2>&1 module load Anaconda3 source $HOME/conda.sh conda activate tf srun python tf.gpu.test.py ``` # Install PyTorch with GPU Login to node650 because it's the AVX512 node and faster. From your login node `srun --gres=gpu:1 --partition=datasci --nodelist=node650 --ntasks=16 --mem=10G --pty bash` Create a new environment (DO NOT LOAD ANY EXISTING ENVIRONMENT) ``` conda create --name torch-cuda python=3.7 conda activate torch-cuda conda install -c "nvidia/label/cuda-11.7.0" cuda-toolkit conda install pytorch torchvision torchaudio pytorch-cuda=11.7 -c pytorch -c nvidia ``` ## Check Torch version `python -c "import torch; print( torch.__version__)"` ## Check CUDA version `python -c "import torch; print(torch. version .cuda)"` ## Check whether Torch is compiled with CUDA `python -c "import torch; print(torch.cuda. is_available())"` # PyTorch3d ``` conda create -n pytorch3d python=3.9 conda activate pytorch3d conda install -c "nvidia/label/cuda-11.7.0" cuda-toolkit conda install pytorch=1.13.0 torchvision torchaudio pytorch-cuda=11.7 -c pytorch -c nvidia conda install -c fvcore -c iopath -c conda-forge fvcore iopath conda install pytorch3d -c pytorch3d ``` # Install picrust2 ``` module load Mamba Miniconda3 mamba create -n picrust2 -c bioconda -c conda-forge picrust2=2.5.1 ``` Once PICRUSt2 is installed, create a file .miniconda3.sh in your $HOME directory and add the following ``` >>> conda initialize >>> !! Contents within this block are managed by 'conda init' !! __conda_setup="$('conda' 'shell.bash' 'hook' 2> /dev/null)" if [ $? -eq 0 ]; then eval "$__conda_setup" else if [ -f "$EBROOTMINICONDA3/etc/profile.d/conda.sh" ]; then . "$EBROOTMINICONDA3/etc/profile.d/conda.sh" else export PATH="$EBROOTMINICONDA3/bin:$PATH" fi fi unset __conda_setup # <<< conda initialize <<< ``` save the file and use `source ~/.miniconda3.sh` conda activate picrust2 PICRUSt2 is now installed, and you can check the installation from the tutorial https://github.com/picrust/picrust2/wiki/PICRUSt2-Tutorial-(v2.5.0) Next time when you use PICRUSt2 always load the Miniconda3 module, source the .miniconda3.sh, and use conda activate to use picrust2. So the steps will be ``` module load Miniconda3 source ~/.miniconda3.sh conda activate picrust2 ``` # Ovito install `conda create -n ovito --strict-channel-priority -c https://conda.ovito.org -c conda-forge ovito`