GPU.md

Docker GPU

To run CLI commands in a Docker container with GPU support out of the box, this project provides the script run_gpu_model.sh. This script builds an image with the project code and runs the image with a selected CLI command.

Step 1: set the GPU id in the config

The config file defines a dictionary to select the id of the GPU to be used:

PROCESSOR_SETTINGS: dict = {
        # "processor_type": "GPU",   # GPU or CPU
        "n_CPU": 16,  # From 0 to 16.
        "GPU_id": None,  # GPU_id = 0 or 1 or None
        "GPU_mb_memory": 38_000,  # Amount of GPU memory in MB
    }

To use CPU, "GPU_id" should be set to None. To use GPU, "GPU_id" should match the desired ID of the GPU. The GPU availables in a server and their IDs can be displayed with the command nvidia-smi.

Step 2: run the gpu script

To run the current project in a detached container, use this command:

./run_gpu_model.sh [OPTIONS] COMMAND [ARGS]...

The same commands defined in the CLI README are available through GPU. In general, changing python src/main.py to ./run_gpu_model.sh for any command will cause the command to run in a GPU container.

Useful commands

To manage the running Docker containers, a useful software is Portainer. It is a UI that allows the user to run, see the logs and stop containers from a browser. To run it:

./portainer.sh

You can check the usage of GPU's and CPU's by your processes by running the following commands in any terminal inside the server:

# to check GPU usage
watch -n 1 nvidia-smi

# to check CPU usage
htop