TemplatesCustom Docker

Custom Docker Template

Run any Docker image on GPU instances.

Launch with Custom Image

curl -X POST https://api.pulserun.dev/v1/instances \
  -H "Authorization: Bearer pr_live_xxxxxxxxxxxxx" \
  -H "Content-Type: application/json" \
  -d '{
    "gpu": "a100_80gb",
    "template": "custom",
    "docker_image": "your-registry/your-image:tag",
    "disk_size_gb": 100
  }'

Port Configuration

By default, port 22 (SSH) is always exposed. Specify additional ports in your Docker image’s EXPOSE directives.

Requirements

  • Image must be publicly accessible (or use a private registry with credentials)
  • NVIDIA GPU drivers are pre-installed on the host
  • Use nvidia/cuda base images for GPU support

Example: Custom Training Image

FROM pytorch/pytorch:2.3.0-cuda12.1-cudnn8-devel
 
WORKDIR /workspace
COPY requirements.txt .
RUN pip install -r requirements.txt
COPY . .
 
CMD ["python", "train.py"]
curl -X POST https://api.pulserun.dev/v1/instances \
  -H "Authorization: Bearer pr_live_xxxxxxxxxxxxx" \
  -H "Content-Type: application/json" \
  -d '{
    "gpu": "a100_80gb",
    "template": "custom",
    "docker_image": "ghcr.io/yourorg/training:latest"
  }'
TensorRT-LLMAutopilot Mode