Enabling NVIDIA GPU support for my DAGs through astro project local

[shayan-ekramnia]

Are there any best practices for enabling NVIDIA GPU support in a local Astro Airflow project (using the Astro Runtime image) so that my DAGs can detect and leverage GPUs for model training? Any example Dockerfile snippets, Compose overrides, or DAG patterns would be greatly appreciated.

I am developing the project on Linux Ubuntu 24.04.2 using Docker Desktop and an NVIDIA L40 GPU and want to train models using torch