Skip to content

Quick Start

Here we show the installation of gRASPA on clusters such as Quest @ Northwestern and on Permutter @ LBNL

Detailed installation process is documentedInstall.md for local machines such as for Ubuntu 24.04 and CentOS 7

🤖 Installation

  • 📝 NOTE: If you don't need deep potential, cd src_clean/, and start from step 6.

  • 📝 NOTE: Check for the compute-capacity (cc) for the GPU, it is related to the architecture of the GPU.

    You need this to correctly compile the code! Here are some cc's for popular cards:

    • Tesla A100: cc80
    • RTX 30 series: cc86
    • RTX 40 series: cc89
    • 📝 use -target=gpu as an easy option in the compilation script

🎤 Step 1

We download TensorFlow2 C++ API to a local directory: (assuming in the HOME directory) 

mkdir ctensorflow
cd ctensorflow
wget https://storage.googleapis.com/tensorflow/libtensorflow/libtensorflow-gpu-linux-x86_64-2.11.0.tar.gz
tar -xvf libtensorflow-gpu-linux-x86_64-2.11.0.tar.gz
cd ..
vi .bashrc

💃 Step 2

Add the following directories to environment variables in the .bashrc file: 

export LIBRARY_PATH=$LIBRARY_PATH:~/ctensorflow/lib
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:~/ctensorflow/lib

🎸 Step 3

Then, we install CppFlow: 

git clone https://github.com/serizba/cppflow
cd cppflow
mkdir build
cd build
cmake -DCMAKE_PREFIX_PATH=~/ctensorflow/ ..
make install DESTDIR=~/ctensorflow/

🏀 Step 4

NERSC has its own PyTorch/LibTorch module, on Quest and other local machines, we recommend libtorch-2.0.1-cu117, please check for LibTorch @ pytorch.org

so now we can start patching gRASPA code with ML potential functionality. For example, if we want to use Allegro model which uses LibTorch: 

mkdir patch_Allegro

🎶 Step 5

Modify line 64 in patch.py file to patch_model=['Allegro']. Then, 

python patch.py
Now the patched source code is in patch_Allegro/. Go into the patched folder, and we have some final work to do. 
cd patch_Allegro/

🐔 Step 6

Finally, we need to modify the source code due to NERSC configuration: 

sed -i "s/std::filesystem/std::experimental::filesystem/g" *
sed -i "s/<filesystem>/<experimental\/filesystem>/g" *

🦋 Step 7

Then, copy NVC_COMPILE_NERSC to the source code folder, and compile the code in the folder as follows:

  • 📝 NOTE: If you use the vanilla version, simply replace NVC_COMPILE_NERSC with NVC_COMPILE_NERSC_VANILLA and proceed. 
    chmod +x NVC_COMPILE_NERSC
./NVC_COMPILE_NERSC
    Remeber to change cppflowDir and tfDir directories in NVC_COMPILE_NERSC if you install CppFlow and TensorFlow API in different directories. You might see some warning messages during compilation, just ignore them. Once ready, you will see a binary excutable nvc_main.x in the folder.

Hooray!!!

GG~ Good work! Now you are ready to go!