Libtorch (Pytorch C++ API)

The way one enables LibTorch Paszke et al. (2019) capabilities in MOOSE depends on the operating system (Linux or Mac) and if we use HPC or just a local workstation.

note

Before we review the main approaches, it is important to emphasize that linking MOOSE with libtorch on Linux machines is not supported if the compiler stack has been built using a libc version below 2.27 (for libtorch v 1.8+) or 2.23 (for libtorch v1.4-1.8). Furthermore, we do not support libtorch versions below v1.4. To check your currently used libc version on Linux machine, use the following command:


ldd --version

Setting up the environment

For Mac workstations, the user needs to create a conda environment using the instructions here. On Linux machines, however, we cannot use the conda packages due to the mismatch between libc versions. For this reason, given that the system libc version allows the linking between the two libraries, we need to install PETSc and libmesh manually. For instructions on the installation of these, see HPC Cluster.

In situations when the libc version allows the linking but the compiler stack has been compiled with an older libc version (HPC machines potentially), we need to build the compiler from scratch. For instructions in this process, visit GCC from Source

Installing Libtorch

The user can choose from two alternatives when it comes to installing libtorch:

Install using the script provided in MOOSE:
For this, navigate to the MOOSE root directory and execute the following script:
```
./scripts/setup_libtorch.sh
```
which downloads libtorch from the official site and sets it up in the framework\contrib directory. The script checks for operating system and libc version and throws errors if the system is not suitable for the coupling.
note
The desired version of libtorch can be set by the following argument:
```
./scripts/setup_libtorch.sh --version=1.8
```
Note that we do not support libtorch below a version of 1.4. The default version downloaded by the script is 1.10.
Install from source: The user can find instructions on how to install libtorch from source on the official website.

Configure and compile MOOSE with libtorch

To achieve this, first configure MOOSE with libtorch support (along with any other desired configure options)


./configure --with-libtorch

note

If you would like to use a custom libtorch distribution or a manually compiled version, the destination of the libtorch directory can be supplied to the configure script by


./configure --with-libtorch=/path/to/my/custom/libtorch

Also note that the library files (.so/.dylib) and the headers in the custom libtorch installation should follow that of the official distribution handled by the setup_libtorch.sh script.

The last step is to compile MOOSE with libtorch support:


make -j 8

After configured, compile MOOSE as normal.

note

Due to namespace conflicts between MOOSE and libtorch, the folders of source files containing libtorch includes need to be excluded from the unit build directories. The makefile of the Stochastic Tools Module serves as a good example on how to achieve this for applications.

References

Adam Paszke, Sam Gross, Francisco Massa, Adam Lerer, James Bradbury, Gregory Chanan, Trevor Killeen, Zeming Lin, Natalia Gimelshein, Luca Antiga, Alban Desmaison, Andreas Kopf, Edward Yang, Zachary DeVito, Martin Raison, Alykhan Tejani, Sasank Chilamkurthy, Benoit Steiner, Lu Fang, Junjie Bai, and Soumith Chintala. Pytorch: an imperative style, high-performance deep learning library. In H. Wallach, H. Larochelle, A. Beygelzimer, F. d\textquotesingle Alché-Buc, E. Fox, and R. Garnett, editors, Advances in Neural Information Processing Systems 32, pages 8024–8035. Curran Associates, Inc., 2019. URL: http://papers.neurips.cc/paper/9015-pytorch-an-imperative-style-high-performance-deep-learning-library.pdf.

@incollection{paszke2019pytorch,
    author = "Paszke, Adam and Gross, Sam and Massa, Francisco and Lerer, Adam and Bradbury, James and Chanan, Gregory and Killeen, Trevor and Lin, Zeming and Gimelshein, Natalia and Antiga, Luca and Desmaison, Alban and Kopf, Andreas and Yang, Edward and DeVito, Zachary and Raison, Martin and Tejani, Alykhan and Chilamkurthy, Sasank and Steiner, Benoit and Fang, Lu and Bai, Junjie and Chintala, Soumith",
    editor = "Wallach, H. and Larochelle, H. and Beygelzimer, A. and d\textquotesingle Alch\'{e}-Buc, F. and Fox, E. and Garnett, R.",
    title = "PyTorch: An Imperative Style, High-Performance Deep Learning Library",
    booktitle = "Advances in Neural Information Processing Systems 32",
    pages = "8024--8035",
    year = "2019",
    publisher = "Curran Associates, Inc.",
    url = "http://papers.neurips.cc/paper/9015-pytorch-an-imperative-style-high-performance-deep-learning-library.pdf"
}

(modules/stochastic_tools/stochastic_tools.mk)

ifeq ($(ENABLE_LIBTORCH),true)
  libtorch_dirs = $(shell find src/libtorch -type d -not -path '*/.libs*' 2> /dev/null)
  converted_dirs = $(foreach i, $(libtorch_dirs), %$(i))
  app_non_unity_dirs = $(converted_dirs)
endif

Setting up the environment
Installing Libtorch
Configure and compile MOOSE with libtorch
References