Tuning Parameters

Use the Quickstart Guide to launch the default NIM. For model-specific setup, refer to Custom Models.

Configuration Options

For the complete configuration reference and all environment variables, refer to Configuration and Environment Variables.

Example Configurations

The following examples highlight runtime options that affect optimization behavior.

TensorNet With Cell Optimization

Run the following commands:

export ALCHEMI_MODEL_PATH=/path/to/model-dir
docker run --rm -ti --name alchemi-bgr --gpus=all \
    -p 8000:8000 --shm-size=8g \
    -v "${ALCHEMI_MODEL_PATH}:/opt/nim/.cache/TensorNet:ro" \
    -e NIM_DISABLE_MODEL_DOWNLOAD=true \
    -e ALCHEMI_NIM_MODEL_TYPE="tensornet" \
    -e ALCHEMI_NIM_MODEL_PATH="/opt/nim/.cache/TensorNet" \
    -e ALCHEMI_NIM_BGR_CELLOPT=true \
    -e ALCHEMI_NIM_BGR_OPTTOL_PRESSURE=0.1 \
    nvcr.io/nim/nvidia/alchemi-bgr:${__container_version}

MACE With DFT-D3 Dispersion

Run the following commands:

export NGC_API_KEY=<Your NGC API Key>
docker run --rm -ti --name alchemi-bgr --gpus=all \
    -e NGC_API_KEY \
    -p 8000:8000 --shm-size=8g \
    -e ALCHEMI_NIM_DFT3_ENABLED=true \
    -e ALCHEMI_NIM_DFT3_PARAM='{"s8": 0.3908, "a1": 0.5660, "a2": 3.1280}' \
    nvcr.io/nim/nvidia/alchemi-bgr:${__container_version}

Per-Request Parameters

Individual requests can override global configuration parameters as shown in the following example:

import requests

package = {
    'atoms': [{
        'coord': [0.0, 0.0, 0.0, 0.0, 0.0, 1.0],
        'numbers': [1, 1],
        'cell': [10, 0, 0, 0, 10, 0, 0, 0, 10],
    }],
    # Override global settings for this request
    'opttol': 0.001,        # Tighter force convergence
    'cellopt': True,        # Enable cell optimization
    'opttol_pressure': 0.1  # Tighter pressure convergence
}

result = requests.post("http://localhost:8000/v1/infer", json=package)
print(result.json())