lorem-jax 0.1.0__tar.gz

lorem_jax-0.1.0/.claude/commands/release.md

@@ -0,0 +1,48 @@
+Create a new release of lorem-jax on PyPI. Optional argument: version bump level (`patch`, `minor`, or `major`).
+
+## Steps
+
+### 1. Pre-flight checks
+
+- Work from a clean checkout of `main` at `origin/main`. If the current checkout is on another branch or has changes, do not switch it; create a worktree instead: `git fetch origin main && git worktree add <scratch>/release origin/main`, and run everything below in there. Remove the worktree at the end.
+- Check that CI is passing on the latest commit of `main`: `gh run list --repo lab-cosmo/lorem-jax --branch main --limit 3`
+- If CI is not green, do NOT proceed — investigate and fix first
+- Check that `pyproject.toml` has no direct-URL dependencies (`@ git+...`). PyPI rejects them, and the publish job would fail after the tag is already pushed. In particular `jax-pme` must be a versioned PyPI dependency.
+
+### 2. Run full local verification
+
+- Run `uvx tox -e lint` — must pass
+- Run `uvx tox -e tests` — must pass
+- Run `uvx tox -e examples` — must pass (calculator, train-mlp, train-finetune, train-bec)
+- Do NOT proceed if any of these fail
+
+### 3. Determine version
+
+- Find the latest git tag with `git describe --tags --abbrev=0` (or note if there are no tags yet; the first release is then `0.1.0` unless the user says otherwise)
+- If a bump level was given ($ARGUMENTS), compute the new version following semver (e.g., `0.1.0` → `0.1.1` for patch, `0.2.0` for minor, `1.0.0` for major)
+- If NO bump level was given, review the changes (step 4) first, then discuss with the user what the appropriate level should be based on the nature of the changes (breaking → major, new features → minor, fixes/maintenance → patch)
+- Confirm the new version with the user before proceeding
+
+### 4. Review changes and write changelog
+
+- Run `git log <last-tag>..HEAD --oneline` to see all commits since the last release (or all commits if no prior tag)
+- Write a brief changelog summarising the changes, grouped by category where appropriate (features, fixes, breaking changes, maintenance, etc.)
+- Present the changelog to the user for review and approval
+
+### 5. Tag and push
+
+- Create an annotated tag: `git tag -a v<version> -m "Release v<version>"`
+- Push the tag: `git push origin v<version>`
+- This triggers `.github/workflows/release.yml`, which builds the package and publishes it to PyPI via trusted publishing from the `release` environment
+- Watch it: `gh run watch` (or `gh run list --repo lab-cosmo/lorem-jax --workflow release.yml --limit 1`), and confirm the publish job succeeded before continuing
+
+### 6. Create GitHub release
+
+- Use `gh release create v<version> --repo lab-cosmo/lorem-jax --title "v<version>" --notes "<changelog>"` to create a GitHub release with the changelog from step 4
+- Confirm the new version is visible on PyPI: `curl -s https://pypi.org/pypi/lorem-jax/json | python -c "import json,sys; print(json.load(sys.stdin)['info']['version'])"`
+
+## Notes
+
+- The version is derived from git tags by `setuptools_scm` (written to `src/lorem/_version.py`, which is gitignored). No files need to be modified for a release.
+- The release workflow uses PyPI trusted publishing (no tokens). The one-time setup, already done, is a `release` environment on the GitHub repo and a trusted publisher for `lorem-jax` on pypi.org pointing at `lab-cosmo/lorem-jax`, `release.yml`, environment `release`.
+- Sandbox: `git fetch`, `git push`, and `gh` must run as bare commands in their own Bash call, with the shell already inside the repo (use a separate `cd` call first). Wrapping them in `cd … &&`, `$(…)`, `;` or a pipe keeps them inside the sandbox, where ssh keys and the gh token are unreachable.

lorem_jax-0.1.0/.github/dependabot.yml

@@ -0,0 +1,11 @@
+version: 2
+updates:
+  - package-ecosystem: "github-actions"
+    directory: "/"
+    schedule:
+      interval: monthly
+    open-pull-requests-limit: 1
+    groups:
+      action-dependencies:
+        patterns:
+          - "*" # A wildcard to create one PR for all dependencies in the ecosystem

lorem_jax-0.1.0/.github/workflows/lint.yml

@@ -0,0 +1,19 @@
+name: Lint
+
+on:
+  pull_request:
+    branches: [main]
+
+jobs:
+  lint:
+    runs-on: ubuntu-latest
+
+    steps:
+    - uses: actions/checkout@v7
+    - name: Set up Python
+      uses: actions/setup-python@v7
+      with:
+        python-version: "3.13"
+    - run: python -m pip install tox
+    - name: Lint the code
+      run: tox -e lint

lorem_jax-0.1.0/.github/workflows/release.yml

@@ -0,0 +1,40 @@
+name: Release
+
+on:
+  push:
+    tags:
+      - "v*"
+
+jobs:
+  build:
+    runs-on: ubuntu-22.04
+
+    steps:
+    - uses: actions/checkout@v7
+      with:
+        fetch-depth: 0
+    - name: Set up Python
+      uses: actions/setup-python@v7
+      with:
+        python-version: "3.13"
+    - run: python -m pip install build
+    - name: Build package
+      run: python -m build
+    - uses: actions/upload-artifact@v7
+      with:
+        name: dist
+        path: dist/
+
+  publish:
+    needs: build
+    runs-on: ubuntu-22.04
+    environment: release
+    permissions:
+      id-token: write
+
+    steps:
+    - uses: actions/download-artifact@v8
+      with:
+        name: dist
+        path: dist/
+    - uses: pypa/gh-action-pypi-publish@release/v1

lorem_jax-0.1.0/.github/workflows/tests.yml

@@ -0,0 +1,47 @@
+name: Tests
+
+on:
+  push:
+    branches: [main]
+  pull_request:
+    # Check all PR
+
+jobs:
+  tests:
+    runs-on: ${{ matrix.os }}
+    strategy:
+      matrix:
+        include:
+          - os: ubuntu-22.04
+            python-version: "3.11"
+          - os: ubuntu-22.04
+            python-version: "3.14"
+          - os: macos-14
+            python-version: "3.14"
+
+    steps:
+    - uses: actions/checkout@v7
+      with:
+        fetch-depth: 0
+    - name: Set up Python ${{ matrix.python-version }}
+      uses: actions/setup-python@v7
+      with:
+        python-version: ${{ matrix.python-version }}
+    - run: python -m pip install tox coverage[toml]
+    - name: run Python tests
+      run: tox -e tests
+
+  examples:
+    runs-on: ubuntu-22.04
+
+    steps:
+    - uses: actions/checkout@v7
+      with:
+        fetch-depth: 0
+    - name: Set up Python
+      uses: actions/setup-python@v7
+      with:
+        python-version: "3.13"
+    - run: python -m pip install tox
+    - name: run examples
+      run: tox -e examples

lorem_jax-0.1.0/.gitignore

@@ -0,0 +1,18 @@
+*.pyc
+*.ipynb_checkpoints*
+__pycache__
+*.egg-info
+*.swp
+*.swo
+*DS_Store
+*coverage*
+_version.py
+*.gz
+
+.tox/
+build/
+dist/
+uv.lock
+
+# training run outputs (train.py defaults workdir to ./run)
+run/

lorem_jax-0.1.0/LICENSE

@@ -0,0 +1,28 @@
+BSD 3-Clause License
+
+Copyright (c) 2026, lorem-jax developers
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+
+1. Redistributions of source code must retain the above copyright notice, this
+   list of conditions and the following disclaimer.
+
+2. Redistributions in binary form must reproduce the above copyright notice,
+   this list of conditions and the following disclaimer in the documentation
+   and/or other materials provided with the distribution.
+
+3. Neither the name of the copyright holder nor the names of its
+   contributors may be used to endorse or promote products derived from
+   this software without specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE
+FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

lorem_jax-0.1.0/PKG-INFO

@@ -0,0 +1,226 @@
+Metadata-Version: 2.4
+Name: lorem-jax
+Version: 0.1.0
+Summary: LOREM: an equivariant long-range machine learning interatomic potential in JAX
+License-Expression: BSD-3-Clause
+Project-URL: Repository, https://github.com/lab-cosmo/lorem-jax
+Requires-Python: >=3.11
+Description-Content-Type: text/markdown
+License-File: LICENSE
+Requires-Dist: ase
+Requires-Dist: e3x
+Requires-Dist: flax
+Requires-Dist: ipi
+Requires-Dist: jax
+Requires-Dist: jax-pme>=0.1
+Requires-Dist: jaxtyping
+Requires-Dist: marathon-train[grain]
+Requires-Dist: grain<0.2.17
+Requires-Dist: opsis
+Requires-Dist: optax
+Dynamic: license-file
+
+# LOREM-JAX
+
+JAX implementation of [LOREM](https://openreview.net/forum?id=pZI9e4SW9P) (Learning Long-Range Representations with Equivariant Messages), a machine learning interatomic potential with equivariant long-range message passing.
+
+Built on [JAX](https://github.com/jax-ml/jax), [Flax](https://github.com/google/flax), [e3x](https://github.com/google-research/e3x), and [jax-pme](https://github.com/lab-cosmo/jax-pme).
+
+## Installation
+
+Requires Python >= 3.11.
+
+```bash
+pip install .
+```
+
+## Usage
+
+### ASE calculator
+
+```python
+import jax
+from ase.build import bulk
+from lorem.models.mlip import Lorem
+from lorem.calculator import Calculator
+
+model = Lorem(cutoff=5.0)
+params = model.init(jax.random.key(42), *model.dummy_inputs())
+calc = Calculator.from_model(model, params=params)
+
+atoms = bulk("Ar") * [2, 2, 2]
+calc.calculate(atoms)
+print(calc.results["energy"], calc.results["forces"].shape)
+```
+
+To load a trained model from a checkpoint:
+
+```python
+calc = Calculator.from_checkpoint("path/to/checkpoint")
+```
+
+### Training
+
+Training a model involves three steps: preparing the data, configuring the model and training settings, and running the training script.
+
+#### 1. Prepare data
+
+Training data is stored in [marathon](https://github.com/sirmarcel/marathon) format. Convert your extended XYZ dataset using a preparation script (see `examples/train-mlp/prepare.py` for a template):
+
+```python
+from marathon.data import datasets, get_splits
+from marathon.grain import prepare
+
+# datasets is a Path resolved from the $DATASETS environment variable
+prepare(train_atoms, folder=datasets / "my_project/train", ...)
+prepare(valid_atoms, folder=datasets / "my_project/valid", ...)
+```
+
+The `$DATASETS` environment variable sets the root directory where prepared datasets are stored. All dataset paths in `settings.yaml` are resolved relative to this directory.
+
+#### 2. Configure the experiment
+
+Each experiment lives in its own directory containing two YAML files:
+
+**`model.yaml`** defines the model architecture:
+
+```yaml
+model:
+  lorem.Lorem:
+    cutoff: 5.0
+    max_degree: 4
+    max_degree_lr: 2
+    num_features: 128
+    num_spherical_features: 4
+    num_message_passing: 1
+```
+
+Use `lorem.LoremBEC` instead of `lorem.Lorem` to train a model that additionally predicts Born effective charges.
+
+**`settings.yaml`** configures training:
+
+```yaml
+train: "my_project/train"           # path relative to $DATASETS
+valid: "my_project/valid"           # path relative to $DATASETS
+seed: 23
+batcher:
+  batch_size: 4
+loss_weights: {"energy": 0.5, "forces": 0.5}
+optimizer: adam                      # adam or muon
+start_learning_rate: 1e-3
+min_learning_rate: 1e-6
+max_epochs: 2000
+valid_every_epoch: 2
+decay_style: linear                  # linear, exponential, or warmup_cosine
+use_wandb: True
+```
+
+<details>
+<summary>All training settings</summary>
+
+| Setting | Default | Description |
+|---|---|---|
+| `train` | *required* | Training dataset path (relative to `$DATASETS`) |
+| `valid` | *required* | Validation dataset path (relative to `$DATASETS`) |
+| `test_datasets` | `{}` | Extra test datasets: `{name: [path, save_predictions]}` |
+| `batcher.batch_size` | *required* | Samples per batch |
+| `batcher.size_strategy` | `powers_of_4` | Padding strategy for batch dimensions |
+| `loss_weights` | `{"energy": 0.5, "forces": 0.5}` | Per-target loss weights |
+| `scale_by_variance` | `False` | Scale loss weights by validation set variance |
+| `optimizer` | `adam` | Optimizer (`adam`, `muon`, or any optax optimizer) |
+| `start_learning_rate` | `1e-3` | Initial learning rate |
+| `min_learning_rate` | `1e-6` | Minimum learning rate |
+| `max_epochs` | `2000` | Maximum training epochs |
+| `valid_every_epoch` | `2` | Validate every N epochs |
+| `decay_style` | `linear` | LR schedule: `linear`, `exponential`, or `warmup_cosine` |
+| `start_decay_after` | `10` | Epoch to begin LR decay |
+| `stop_decay_after` | `max_epochs` | Epoch to end LR decay (linear only) |
+| `warmup_epochs` | `0` | Warmup epochs (`warmup_cosine` only) |
+| `gradient_clip` | `0` | Gradient clipping threshold (0 = disabled) |
+| `seed` | `0` | Random seed |
+| `rotational_augmentation` | `False` | Apply random rotations to training data |
+| `filter_mixed_pbc` | `False` | Filter out structures with mixed periodic boundary conditions |
+| `filter_above_num_atoms` | `False` | Filter out structures above this atom count |
+| `checkpointers` | `default` | `default` or `full` (adds RMSE checkpointers) |
+| `use_wandb` | `True` | Log to Weights & Biases |
+| `wandb_project` | auto | W&B project name (default: derived from folder names) |
+| `wandb_name` | auto | W&B run name (default: experiment folder name) |
+| `benchmark_pipeline` | `True` | Benchmark data pipeline before training |
+| `compilation_cache` | `False` | Enable JAX persistent compilation cache |
+| `default_matmul_precision` | `float32` | JAX matmul precision (`default`, `float32`) |
+| `debug_nans` | `False` | Enable JAX NaN debugging (~50% slowdown) |
+| `enable_x64` | `False` | Enable 64-bit floating point |
+| `worker_count` | `4` | Data loading workers (training) |
+| `worker_count_valid` | `worker_count` | Data loading workers (validation) |
+| `worker_buffer_size` | `2` | Prefetch buffer per worker (training) |
+
+</details>
+
+#### 3. Run training
+
+```bash
+cd my_experiment
+DATASETS=/path/to/datasets lorem-train
+```
+
+Training writes checkpoints, logs, and plots to a `run/` directory inside the experiment folder. If a `run/` directory already exists, training resumes from the latest checkpoint.
+
+See `examples/train-mlp/` and `examples/train-bec/` for complete examples including data preparation and configuration files.
+
+### Model variants
+
+- **`Lorem`** -- the standard MLIP model (energy + forces + stress)
+- **`LoremBEC`** -- predicts Born effective charges in addition to energy/forces
+
+### Key hyperparameters
+
+| Parameter | Default | Description |
+|---|---|---|
+| `cutoff` | 5.0 | Short-range cutoff radius (A) |
+| `max_degree` | 6 | Maximum angular momentum for spherical features |
+| `max_degree_lr` | 2 | Maximum angular momentum for long-range charges |
+| `num_features` | 128 | Number of scalar features |
+| `num_spherical_features` | 8 | Number of spherical feature channels |
+| `num_radial` | 32 | Number of radial basis functions |
+| `num_message_passing` | 0 | Number of short-range message passing steps |
+| `lr` | True | Enable long-range (Ewald) interaction |
+
+## Installing the i-PI driver
+
+After installation of the package, install the i-PI driver via:
+
+```bash
+lorem-install-ipi-driver
+```
+
+This copies the LOREM driver into the i-PI `pes` directory. You can rerun `lorem-install-ipi-driver` anytime (it is idempotent) if you switch environments or reinstall i-PI.
+
+## Development
+
+Format and lint:
+
+```bash
+ruff format . && ruff check --fix .
+```
+
+Run tests:
+
+```bash
+python -m pytest tests/ -v --override-ini="addopts="
+```
+
+Or use tox:
+
+```bash
+tox -e lint       # check formatting + linting
+tox -e tests      # run unit tests
+tox -e examples   # run examples as smoke tests
+tox -e format     # auto-format
+```
+
+## License
+
+BSD-3-Clause
+
+
+This project is [maintained](https://github.com/lab-cosmo/.github/blob/main/Maintainers.md) by [@PicoCentauri](https://github.com/PicoCentauri) and [@sirmarcel](https://github.com/sirmarcel), who will reply to issues and pull requests opened on this repository as soon as possible. You can mention them directly if you did not receive an answer after a couple of days.

lorem_jax-0.1.0/README.md

@@ -0,0 +1,204 @@
+# LOREM-JAX
+
+JAX implementation of [LOREM](https://openreview.net/forum?id=pZI9e4SW9P) (Learning Long-Range Representations with Equivariant Messages), a machine learning interatomic potential with equivariant long-range message passing.
+
+Built on [JAX](https://github.com/jax-ml/jax), [Flax](https://github.com/google/flax), [e3x](https://github.com/google-research/e3x), and [jax-pme](https://github.com/lab-cosmo/jax-pme).
+
+## Installation
+
+Requires Python >= 3.11.
+
+```bash
+pip install .
+```
+
+## Usage
+
+### ASE calculator
+
+```python
+import jax
+from ase.build import bulk
+from lorem.models.mlip import Lorem
+from lorem.calculator import Calculator
+
+model = Lorem(cutoff=5.0)
+params = model.init(jax.random.key(42), *model.dummy_inputs())
+calc = Calculator.from_model(model, params=params)
+
+atoms = bulk("Ar") * [2, 2, 2]
+calc.calculate(atoms)
+print(calc.results["energy"], calc.results["forces"].shape)
+```
+
+To load a trained model from a checkpoint:
+
+```python
+calc = Calculator.from_checkpoint("path/to/checkpoint")
+```
+
+### Training
+
+Training a model involves three steps: preparing the data, configuring the model and training settings, and running the training script.
+
+#### 1. Prepare data
+
+Training data is stored in [marathon](https://github.com/sirmarcel/marathon) format. Convert your extended XYZ dataset using a preparation script (see `examples/train-mlp/prepare.py` for a template):
+
+```python
+from marathon.data import datasets, get_splits
+from marathon.grain import prepare
+
+# datasets is a Path resolved from the $DATASETS environment variable
+prepare(train_atoms, folder=datasets / "my_project/train", ...)
+prepare(valid_atoms, folder=datasets / "my_project/valid", ...)
+```
+
+The `$DATASETS` environment variable sets the root directory where prepared datasets are stored. All dataset paths in `settings.yaml` are resolved relative to this directory.
+
+#### 2. Configure the experiment
+
+Each experiment lives in its own directory containing two YAML files:
+
+**`model.yaml`** defines the model architecture:
+
+```yaml
+model:
+  lorem.Lorem:
+    cutoff: 5.0
+    max_degree: 4
+    max_degree_lr: 2
+    num_features: 128
+    num_spherical_features: 4
+    num_message_passing: 1
+```
+
+Use `lorem.LoremBEC` instead of `lorem.Lorem` to train a model that additionally predicts Born effective charges.
+
+**`settings.yaml`** configures training:
+
+```yaml
+train: "my_project/train"           # path relative to $DATASETS
+valid: "my_project/valid"           # path relative to $DATASETS
+seed: 23
+batcher:
+  batch_size: 4
+loss_weights: {"energy": 0.5, "forces": 0.5}
+optimizer: adam                      # adam or muon
+start_learning_rate: 1e-3
+min_learning_rate: 1e-6
+max_epochs: 2000
+valid_every_epoch: 2
+decay_style: linear                  # linear, exponential, or warmup_cosine
+use_wandb: True
+```
+
+<details>
+<summary>All training settings</summary>
+
+| Setting | Default | Description |
+|---|---|---|
+| `train` | *required* | Training dataset path (relative to `$DATASETS`) |
+| `valid` | *required* | Validation dataset path (relative to `$DATASETS`) |
+| `test_datasets` | `{}` | Extra test datasets: `{name: [path, save_predictions]}` |
+| `batcher.batch_size` | *required* | Samples per batch |
+| `batcher.size_strategy` | `powers_of_4` | Padding strategy for batch dimensions |
+| `loss_weights` | `{"energy": 0.5, "forces": 0.5}` | Per-target loss weights |
+| `scale_by_variance` | `False` | Scale loss weights by validation set variance |
+| `optimizer` | `adam` | Optimizer (`adam`, `muon`, or any optax optimizer) |
+| `start_learning_rate` | `1e-3` | Initial learning rate |
+| `min_learning_rate` | `1e-6` | Minimum learning rate |
+| `max_epochs` | `2000` | Maximum training epochs |
+| `valid_every_epoch` | `2` | Validate every N epochs |
+| `decay_style` | `linear` | LR schedule: `linear`, `exponential`, or `warmup_cosine` |
+| `start_decay_after` | `10` | Epoch to begin LR decay |
+| `stop_decay_after` | `max_epochs` | Epoch to end LR decay (linear only) |
+| `warmup_epochs` | `0` | Warmup epochs (`warmup_cosine` only) |
+| `gradient_clip` | `0` | Gradient clipping threshold (0 = disabled) |
+| `seed` | `0` | Random seed |
+| `rotational_augmentation` | `False` | Apply random rotations to training data |
+| `filter_mixed_pbc` | `False` | Filter out structures with mixed periodic boundary conditions |
+| `filter_above_num_atoms` | `False` | Filter out structures above this atom count |
+| `checkpointers` | `default` | `default` or `full` (adds RMSE checkpointers) |
+| `use_wandb` | `True` | Log to Weights & Biases |
+| `wandb_project` | auto | W&B project name (default: derived from folder names) |
+| `wandb_name` | auto | W&B run name (default: experiment folder name) |
+| `benchmark_pipeline` | `True` | Benchmark data pipeline before training |
+| `compilation_cache` | `False` | Enable JAX persistent compilation cache |
+| `default_matmul_precision` | `float32` | JAX matmul precision (`default`, `float32`) |
+| `debug_nans` | `False` | Enable JAX NaN debugging (~50% slowdown) |
+| `enable_x64` | `False` | Enable 64-bit floating point |
+| `worker_count` | `4` | Data loading workers (training) |
+| `worker_count_valid` | `worker_count` | Data loading workers (validation) |
+| `worker_buffer_size` | `2` | Prefetch buffer per worker (training) |
+
+</details>
+
+#### 3. Run training
+
+```bash
+cd my_experiment
+DATASETS=/path/to/datasets lorem-train
+```
+
+Training writes checkpoints, logs, and plots to a `run/` directory inside the experiment folder. If a `run/` directory already exists, training resumes from the latest checkpoint.
+
+See `examples/train-mlp/` and `examples/train-bec/` for complete examples including data preparation and configuration files.
+
+### Model variants
+
+- **`Lorem`** -- the standard MLIP model (energy + forces + stress)
+- **`LoremBEC`** -- predicts Born effective charges in addition to energy/forces
+
+### Key hyperparameters
+
+| Parameter | Default | Description |
+|---|---|---|
+| `cutoff` | 5.0 | Short-range cutoff radius (A) |
+| `max_degree` | 6 | Maximum angular momentum for spherical features |
+| `max_degree_lr` | 2 | Maximum angular momentum for long-range charges |
+| `num_features` | 128 | Number of scalar features |
+| `num_spherical_features` | 8 | Number of spherical feature channels |
+| `num_radial` | 32 | Number of radial basis functions |
+| `num_message_passing` | 0 | Number of short-range message passing steps |
+| `lr` | True | Enable long-range (Ewald) interaction |
+
+## Installing the i-PI driver
+
+After installation of the package, install the i-PI driver via:
+
+```bash
+lorem-install-ipi-driver
+```
+
+This copies the LOREM driver into the i-PI `pes` directory. You can rerun `lorem-install-ipi-driver` anytime (it is idempotent) if you switch environments or reinstall i-PI.
+
+## Development
+
+Format and lint:
+
+```bash
+ruff format . && ruff check --fix .
+```
+
+Run tests:
+
+```bash
+python -m pytest tests/ -v --override-ini="addopts="
+```
+
+Or use tox:
+
+```bash
+tox -e lint       # check formatting + linting
+tox -e tests      # run unit tests
+tox -e examples   # run examples as smoke tests
+tox -e format     # auto-format
+```
+
+## License
+
+BSD-3-Clause
+
+
+This project is [maintained](https://github.com/lab-cosmo/.github/blob/main/Maintainers.md) by [@PicoCentauri](https://github.com/PicoCentauri) and [@sirmarcel](https://github.com/sirmarcel), who will reply to issues and pull requests opened on this repository as soon as possible. You can mention them directly if you did not receive an answer after a couple of days.