Templates

Templates are pre-configured container images that define the software environment for your GPU instances and serverless endpoints. Use templates to standardize your deployment stack and speed up launches.

Creating a Template

curl -X POST https://api.gpuworker.com/templates \
  -H "Authorization: Bearer YOUR_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "pytorch-2.1-jupyter",
    "imageName": "nvcr.io/nvidia/pytorch:2.1.0-py3.10-cuda12.1",
    "dockerStartCmd": "bash -c '\''jupyter lab --ip=0.0.0.0 --allow-root --no-browser'\''",
    "containerDiskInGb": 20,
    "volumeInGb": 50
  }'

Template Fields

Field	Description
name	Human-readable template name
imageName	Docker image (e.g., nvcr.io/nvidia/pytorch:2.1.0-py3.10-cuda12.1)
dockerStartCmd	Command to run on container start
containerDiskInGb	Size of the container’s local SSD storage
volumeInGb	Size of the attached network volume

Using Private Images

To use images from private container registries:

1. Add your registry credentials in Container Registry
2. Reference the private image in your template’s imageName

Popular Base Images

Image	Description
nvcr.io/nvidia/pytorch:2.1.0-py3.10-cuda12.1	PyTorch 2.1 with CUDA 12.1
stabilityai/stable-diffusion:web-ui	Stable Diffusion WebUI
nvidia/cuda:12.1.0-devel-ubuntu22.04	Base CUDA development

Best Practices

1. Pin image versions — Use specific tags, not latest, for reproducibility
2. Minimize image size — Smaller images mean faster cold starts
3. Pre-install dependencies — Bake your requirements into the image to avoid install-on-start delays
4. Use startup scripts — Set dockerStartCmd to automatically launch services (Jupyter, API servers, etc.)