Docker image (#86)

* Update README.md

* Create Dockerfile

* Update README.md

* Update Dockerfile

* Update README.md

* Update .gitignore

* Update Dockerfile to v3 weights

* Update Dockerfile

Co-authored-by: Ben Rhodes <benjamin.rhodes26@gmail.com>

* Fix docker image and readme

---------

Co-authored-by: Colby Ford <colbytylerford@gmail.com>
Co-authored-by: Colby T. Ford <colby@tuple.xyz>
Co-authored-by: ben rhodes <benrhodes@bens-MacBook-Pro.local>

Author: 3 people

.gitignore

@@ -162,4 +162,7 @@ cython_debug/
 #.idea/
 
 # Private
-datasets/
+datasets/
+
+# VS Code
+.devcontainer/

Dockerfile

@@ -0,0 +1,23 @@
+FROM pytorch/pytorch:2.6.0-cuda12.4-cudnn9-runtime
+
+ENV DEBIAN_FRONTEND=noninteractive
+
+## Install system requirements
+RUN apt-get update && \
+    apt-get install -y \
+        ca-certificates \
+        wget \
+        git \
+        sudo \
+        gcc \
+        g++
+
+# Help Numba find lubcudart.so
+ENV LD_LIBRARY_PATH=/opt/conda/lib/python3.11/site-packages/nvidia/cuda_runtime/lib:$LD_LIBRARY_PATH
+RUN ln -s \
+    /opt/conda/lib/python3.11/site-packages/nvidia/cuda_runtime/lib/libcudart.so.12 \
+    /opt/conda/lib/python3.11/site-packages/nvidia/cuda_runtime/lib/libcudart.so
+
+## Install Python requirements
+RUN pip install orb-models && \
+    pip install --extra-index-url=https://pypi.nvidia.com "cuml-cu12==25.2.*"

README.md

@@ -23,6 +23,8 @@ pip install --extra-index-url=https://pypi.nvidia.com "cuml-cu11==25.2.*"  # Fo
 pip install --extra-index-url=https://pypi.nvidia.com "cuml-cu12==25.2.*"  # For cuda versions >=12.0, <13.0
 ```
 
+Alternatively, you can use Docker to run orb-models; [see instructions below](#docker).
+
 ### Updates
 
 **April 2025**: We have released the [Orb-v3 set of potentials](https://arxiv.org/abs/2504.06231). These models improve substantially over Orb-v2, in particular:
@@ -244,6 +246,20 @@ model = getattr(pretrained, <base_model>)(
 > - The script only tracks a limited set of metrics (energy/force/stress MAEs) which may be insufficient for some downstream use-cases. For instance, if you wish to finetune a model for Molecular Dynamics simulations, we have found (anecdotally) that models that are just on the cusp of overfitting to force MAEs can be substantially worse for simulations. Ideally, more robust "rollout" metrics would be included in the finetuning training loop. In lieu of this, we recommend more aggressive early-stopping i.e. using models several epochs prior to any sign of overfitting.
 
 
+## Docker
+
+You can run orb-models using Docker, which provides a consistent environment with all dependencies pre-installed:
+
+1. Build the Docker image locally:
+
+   ```bash
+   docker build -t orb_models .
+   ```
+2. Run the Docker container:
+
+   ```bash
+   docker run --gpus all --rm --name orb_models -it orb_models /bin/bash
+   ```
 
 
 ### Citing
@@ -282,4 +298,4 @@ ORB models are licensed under the Apache License, Version 2.0. Please see the [L
 
 ### Community
 
-Please join the discussion on Discord by following [this](https://discord.gg/SyD6vWSSTB) link.
+Please join the discussion on Discord by following [this](https://discord.gg/SyD6vWSSTB) link.