This project aims to create a set of Docker images based on the Wolfi Linux distribution, optimized for NVIDIA CUDA and various deep learning frameworks. The goal is to provide a lightweight and efficient base image for building and deploying AI/ML applications.
The following table compares the sizes of base NVIDIA images, NVIDIA CUDA images, and the Wolfi-based images generated by this script:
| Image Type | Size (MB) |
|---|---|
| NVIDIA CUDA Base | 1.4 GB |
| NVIDIA CUDA 12.4.1 | 2.5 GB |
| Wolfi CUDA 12.4.1 (Base) | 382 MB |
| Wolfi CUDA 12.4.1 (TensorFlow) | 1.2 GB |
| Wolfi CUDA 12.4.1 (PyTorch) | 1.1 GB |
As shown in the table, the Wolfi-based images generated by this script are significantly smaller in size compared to the NVIDIA CUDA images, making them ideal for deployment in resource-constrained environments. The size reduction is achieved by using the Wolfi Linux distribution, which is designed to be lightweight and efficient, and by optimizing the installation of CUDA and deep learning frameworks using Micromamba.