tao-run-on-docker

To run Docker containers on an NVIDIA GPU host, use the NVIDIA Container Toolkit with specific NVIDIA driver and CUDA Toolkit versions. This skill provides operational instructions for efficient GPU container deployment using standard docker run conventions.

Prerequisites for running GPU workloads in Docker include a host setup with CUDA Toolkit, NVIDIA drivers, and Docker. You must first configure the GPU host environment using the tao-setup-nvidia-gpu-host skill before starting GPU containers.

Yes, the NVIDIA Container Toolkit supports environment variable passthrough for docker run commands. This skill implements best practices for authentication, mount patterns, and env var passthrough to ensure NVIDIA GPU container workloads run efficiently.

The best way to configure Docker for NVIDIA GPU model deployment is to follow optimized Docker conventions with the NVIDIA Container Toolkit. This ensures compatibility and safe execution of GPU container workloads on supported platforms with specific configurations.

Docker GPU containers fail to start when the host lacks proper NVIDIA Container Toolkit configuration or specific NVIDIA driver and CUDA Toolkit versions. Use this skill to apply correct mount patterns and environment setup for successful GPU container execution.