# Code and Software Submission

## Manuscript Information
- Title: LAMBench: A Benchmark for Large Atomic Models
- Corresponding Author: Anyang Peng, Linfeng Zhang, Han Wang

## System requirements
- All software dependencies and operating systems: described at https://github.com/deepmodeling/LAMBench/blob/main/pyproject.toml
- Versions the software has been tested on: [version 0.1.0](https://github.com/deepmodeling/LAMBench/releases/tag/0.1.0)
- Any required non-standard hardware: None; all required software can be downloaded from PyPI

## Installation guide
- Instructions: please refer to [the readme file](https://github.com/deepmodeling/LAMBench/blob/main/README.md#installation)
- Typical install time on a "normal" desktop computer: 5 minutes

## Demo
- Instructions to run on data:

Below is an example of how to run tests on molecular dynamics (MD) using the Orb-v2 model.

```python
from lambench.models.ase_models import ASEModel
model = ASEModel(
        model_name="orb_v2",
        model_type="ASE",
        model_family="ORB",
        virtualenv="",
        model_metadata={
			"date_added": "2025-04-01",
            "pretty_name":"Orb-v2",
            "num_parameters":25161727,
            "packages": {"orb-models": "0.4.2"}
        }
    )
from lambench.tasks import CalculatorTask
task = CalculatorTask(task_name="nve_md", test_data=None, calculator_params={"num_steps": 100})
# Run 100 steps of MD for demonstration
model.evaluate(task)
```

- Expected output:
```python
{'metrics': {'Gd2Si4Ni2': {'simulation_time': 2.151423215866089, 'steps': 100, 'momenta_diff': np.float64(2.0086446187567793e-07), 'slope': np.float64(0.5676583842486187)}, 'BaTm2NiO5': {'simulation_time': 1.8821752071380615, 'steps': 100, 'momenta_diff': np.float64(1.109865651144807e-07), 'slope': np.float64(0.7703909405165947)}, 'NdPd3': {'simulation_time': 2.0016138553619385, 'steps': 100, 'momenta_diff': np.float64(1.1562258728877268e-07), 'slope': np.float64(0.0419704920791998)}, 'BaV2Ni2O8': {'simulation_time': 2.1841065883636475, 'steps': 100, 'momenta_diff': np.float64(4.4297247082736533e-07), 'slope': np.float64(0.02688047491287327)}, 'SN2CN3H3': {'simulation_time': 2.736257791519165, 'steps': 100, 'momenta_diff': np.float64(5.289080102309365e-07), 'slope': np.float64(1.8391390630221292)}, 'C3N2H5': {'simulation_time': 2.8398561477661133, 'steps': 100, 'momenta_diff': np.float64(3.736058215997301e-07), 'slope': np.float64(2.212501230119175)}, 'H7C4NO': {'simulation_time': 2.739593267440796, 'steps': 100, 'momenta_diff': np.float64(5.933518129756298e-07), 'slope': np.float64(1.0417677363724114)}, 'Cs8N2': {'simulation_time': 2.1909828186035156, 'steps': 100, 'momenta_diff': np.float64(3.072067149871919e-07), 'slope': np.float64(2.6016108074557773)}, 'H2Al32Cr48Mn16N2O': {'simulation_time': 2.357931137084961, 'steps': 100, 'momenta_diff': np.float64(4.2663130754921755e-06), 'slope': np.float64(0.4589563823400662)}}}
```

- Expected run time for demo: 2 minutes with a GPU

## Instructions for use
- How to run the software on your data: Please refer to the "contributing" section of the readme file.
```