Is this page helpful?

Getting Started#

This section describes how to launch GenMol NIM v2.0.0 and submit your first request.

Prerequisites#

Review the Prerequisites and Support Matrix before launching the container.

Launch GenMol NIM#

  1. Authenticate with the NGC container registry.

docker login nvcr.io --username '$oauthtoken' --password $NGC_API_KEY

  1. Pull the NIM container.

docker pull nvcr.io/nim/nvidia/genmol:2.0.0

  1. Run the container with a persistent cache mount.

Note

Set NGC_API_KEY in your shell before launching the container. For more information, refer to the NGC personal API key guide.

export LOCAL_NIM_CACHE=~/.cache/nim/genmol
mkdir -p "$LOCAL_NIM_CACHE"

docker run --rm -it --name genmol-nim \
  --runtime=nvidia --gpus=all \
  -e NVIDIA_VISIBLE_DEVICES=0 \
  --shm-size=2G \
  --ulimit memlock=-1 \
  --ulimit stack=67108864 \
  -e NGC_API_KEY=$NGC_API_KEY \
  -e NIM_HTTP_API_PORT=8000 \
  -v "$LOCAL_NIM_CACHE:/opt/nim/.cache" \
  -p 8000:8000 \
  nvcr.io/nim/nvidia/genmol:2.0.0

  1. Open a new terminal and wait for the service to become ready.

curl localhost:8000/v1/health/ready
...
true

Stop the Container#

docker stop genmol-nim

Run Inference in Linux Shell#

curl --silent --request POST \
  --url 'http://127.0.0.1:8000/generate' \
  --header 'Content-Type: application/json' \
  --output output.json \
  --data '{
    "smiles": "CCS2(=O)(=O).C134CN2C1.C3C#N.[*{15-15}]",
    "num_molecules": 10,
    "temperature": 1.0,
    "noise": 1.0,
    "scoring": "QED"
  }'

Example response:

{
  "status": "success",
  "molecules": [
    {
      "smiles": "CCS(=O)(=O)N1CC(CC#N)(c2ccc(Cl)c(F)c2)C1",
      "score": 0.856
    }
  ]
}

Run Inference with Python#

import json
import requests

session = requests.Session()
response = session.post(
    "http://127.0.0.1:8000/generate",
    headers={"Accept": "application/json"},
    json={
        "smiles": "CCS2(=O)(=O).C134CN2C1.C3C#N.[*{15-15}]",
        "num_molecules": 5,
        "temperature": 1.0,
        "noise": 1.0,
        "scoring": "QED",
    },
)

response.raise_for_status()
print(json.dumps(response.json(), indent=2))

Input and Output Format#

The input is a SAFE text sequence with masked fragments as a template for molecular generation. SAFE extends SMILES with explicit fragment and attachment-point structure. The masked fragment placeholder uses the form [*{min_len-max_len}], where min_len and max_len specify the number of tokens to generate for that region.

If an empty template is provided, the model performs de novo generation. If a partial SAFE or SMILES template is provided, the model performs conditioned generation such as motif extension, scaffold decoration, or linker design.