| Title
A single sentence description.
|
Torch ML model driver v001 |
|---|---|
| Description |
Model driver (MD) for TorchScript based machine learning interatomic potentials (MLIPs). This MD interfaces with libtorch (the pytorch C++ interface), and libdescriptor. Dependencies: 1. Libtorch [essential] 2. torch_scatter [optional for GNNs] 3. torch_sparse [optional for GNNs] 4. libdescriptor [optional for descriptor based models] The dependencies can be installed using the provided script "install_dependencies.sh" (see files below) which installs all the dependencies in the currently working folder. To activate the environment source the generated `env.sh` file (or copy its contents into a `.bashrc` file to autmatically initializing the environment in the future). The MD provides an option for offloading the MLIP computations to GPUs. This option is activated by setting the environment variable `KIM_MODEL_EXECUTION_DEVICE` to "cuda". There are also several compilation options to modify the MD behavior, controlled by setting environmental variables. 1. `KIM_MODEL_MPI_AWARE` - If set to `yes` (*case-sensitive*), during driver installation the model driver will be built with MPI support and will require a valid MPI environment to be present at installation time. This ensures a more hardware agnostic allocation of GPU resources. Specifically, this will set up `n` GPUs on each node to be used for `m` ranks on the same node in a round-robin fashion (i.e. rank `m` will receive GPU number [`m` mod `n`]). 2. `KIM_MODEL_DISBALE_GRAPH` - If this environment variable is defined (irrespective of value), the model driver will be built without graph support. This means that during build time it will not try to find and link against `libtorchscatter` and `libtorchsparse` libraries. Consult the README.md file for fixing commonly installation problems. |
| Disclaimer
A statement of applicability provided by the contributor, informing users of the intended use of this KIM Item.
|
None |
| Contributor |
Amit Gupta |
| Maintainer |
Amit Gupta |
| Developer |
Amit Gupta Daniel S. Karls |
| Published on KIM | 2025 |
| How to Cite |
This Model Driver is archived in OpenKIM [1-3]. [1] Gupta A, Karls DS. Torch ML model driver v001. OpenKIM; 2025. doi:10.25950/aefd2d8a [2] Tadmor EB, Elliott RS, Sethna JP, Miller RE, Becker CA. The potential of atomistic simulations and the Knowledgebase of Interatomic Models. JOM. 2011;63(7):17. doi:10.1007/s11837-011-0102-6 [3] Elliott RS, Tadmor EB. Knowledgebase of Interatomic Models (KIM) Application Programming Interface (API). OpenKIM; 2011. doi:10.25950/ff8f563a Click here to download the above citation in BibTeX format. |
| Funding | Not available |
| Short KIM ID
The unique KIM identifier code.
| MD_173118614730_001 |
| Extended KIM ID
The long form of the KIM ID including a human readable prefix (100 characters max), two underscores, and the Short KIM ID. Extended KIM IDs can only contain alpha-numeric characters (letters and digits) and underscores and must begin with a letter.
| TorchML__MD_173118614730_001 |
| DOI |
10.25950/aefd2d8a https://doi.org/10.25950/aefd2d8a https://commons.datacite.org/doi.org/10.25950/aefd2d8a |
| KIM Item Type | Model Driver |
| KIM API Version | 2.3 |
| Programming Language(s)
The programming languages used in the code and the percentage of the code written in each one.
| 82.68% C++ 16.38% Shell 0.94% C |
| Previous Version | TorchML__MD_173118614730_000 |
| TorchML_MACE_GuptaTadmorMartiniani_2024_Si__MO_781946209112_001 |
| TorchML_NequIP_GuptaTadmorMartiniani_2024_Si__MO_196181738937_001 |
| TorchML__MD_173118614730_001.txz | Tar+XZ | Linux and OS X archive |
| TorchML__MD_173118614730_001.zip | Zip | Windows archive |