With my recent exploration of GenAI and using a private ChatGPT solution with my own blog posts, I quickly realized in the space of AI/ML, the required software dependencies can take up a significant amount of storage, especially for a kubernetes/container-based deployment.
To give you an example, to deploy the private ChatGPT (h2ogpt) application using kubernetes, just the container image itself is a whopping 40GB+! 😲
Unfourntately, this is not a one off scenario but a common theme when working in the AI/ML space that the size of the packages and drivers are extremely large even when using containers. I figure I should probably setup my own container registry instead of pulling directly from the Internet given the size of these images.
I already have a local Harbor instance running in a VM but with my Synology, I have been using it centralize a number of functions and that would be the ideal place to actually run Harbor. While you can run individual containers on the Synology as I have demonstrated HERE with GitLab, the Harbor installation processes relies on Docker Compose, which Synology does not natively support using the Synology DiskStation Manager (DSM) interface.
With a little bit of tinkering and trial/error, I was able to finally get Harbor running on my Synology and centralize all my storage needs including having my own container registry.