add model card: Academic Software License (ASL) + repo/usage docs
Browse files
README.md
ADDED
|
@@ -0,0 +1,58 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: other
|
| 3 |
+
license_name: academic-software-license
|
| 4 |
+
license_link: https://github.com/ICAMS/grace-tensorpotential/blob/master/LICENSE.md
|
| 5 |
+
library_name: tensorpotential
|
| 6 |
+
tags:
|
| 7 |
+
- grace
|
| 8 |
+
- interatomic-potential
|
| 9 |
+
- molecular-dynamics
|
| 10 |
+
- machine-learning-potential
|
| 11 |
+
- materials-science
|
| 12 |
+
---
|
| 13 |
+
|
| 14 |
+
# GRACE foundation models
|
| 15 |
+
|
| 16 |
+
This repository distributes **GRACE** (Graph Atomic Cluster Expansion) machine-learning
|
| 17 |
+
interatomic potentials — foundation models fitted with
|
| 18 |
+
[GRACEmaker / tensorpotential](https://github.com/ICAMS/grace-tensorpotential).
|
| 19 |
+
|
| 20 |
+
## License
|
| 21 |
+
|
| 22 |
+
**All GRACE models distributed in this repository are released under the Academic Software
|
| 23 |
+
License (ASL).** By downloading or using any model from this repository you agree to the terms
|
| 24 |
+
of the ASL: <https://github.com/ICAMS/grace-tensorpotential/blob/master/LICENSE.md>
|
| 25 |
+
|
| 26 |
+
## Contents
|
| 27 |
+
|
| 28 |
+
Each model is provided as two `.tar.gz` archives:
|
| 29 |
+
|
| 30 |
+
- `models/<model>-model.tar.gz` — the deployable TensorFlow `SavedModel` directory
|
| 31 |
+
(used by the ASE `TPCalculator`). UQ-enabled models embed an uncertainty head, and many
|
| 32 |
+
archives also bundle `kokkos.npz` for LAMMPS-Kokkos deployment.
|
| 33 |
+
- `checkpoints/<model>-checkpoint.tar.gz` — the training checkpoint
|
| 34 |
+
(`model.yaml`, `checkpoint.index`, `checkpoint.data-*`, and `gmm_artifacts.npz` for UQ models).
|
| 35 |
+
|
| 36 |
+
## Usage
|
| 37 |
+
|
| 38 |
+
These models are normally fetched automatically by the `tensorpotential` / GRACEmaker
|
| 39 |
+
tooling (`grace_models`, `TPCalculator`, foundation-model presets). They can also be
|
| 40 |
+
downloaded directly, e.g.:
|
| 41 |
+
|
| 42 |
+
```python
|
| 43 |
+
from huggingface_hub import hf_hub_download
|
| 44 |
+
path = hf_hub_download(
|
| 45 |
+
repo_id="AMS-ICAMS-RUB/grace-foundation-models",
|
| 46 |
+
filename="models/GRACE-2L-OMAT-large-base-model.tar.gz",
|
| 47 |
+
revision="kk", # immutable tag; "main" is the moving tip
|
| 48 |
+
)
|
| 49 |
+
```
|
| 50 |
+
|
| 51 |
+
## Versioning
|
| 52 |
+
|
| 53 |
+
Revisions are pinned via immutable git tags so downloads are reproducible:
|
| 54 |
+
|
| 55 |
+
- **`uq-v1`** — the initial set of UQ-augmented OMAT foundation models.
|
| 56 |
+
- **`kk`** — model archives that additionally bundle the LAMMPS-Kokkos export (`kokkos.npz`).
|
| 57 |
+
|
| 58 |
+
`main` always points at the latest revision.
|