Replies: 3 comments 1 reply
-
I don't think this is the right place to ask; that image is maintained by NVIDIA (but they happen to use Docker Hub to distribute the image); from the information on Docker Hub, it looks like the source-repository is https://gitlab.com/nvidia/cuda, so that may be the best location to ask the maintainers of that image; |
Beta Was this translation helpful? Give feedback.
-
@thaJeztah We are from the PyTorch team and looking for ways to improve the docker footprint for PyTorch and related libraries installs. We would like to partner with Docker to find a better solution. Current images from Nvidia have lot of bloat making it challenging to use esp for Inference workloads. Every customer has to go through the pain point of creating optimized images for their environment. Are there any other suggestions from Docker development team that we can explore? Will be happy to discuss more in a meeting |
Beta Was this translation helpful? Give feedback.
-
If you're interested in working with Docker on base images, the place to ask/make suggestions is probably docker-library/official-images. That being said, generally images there are required to be free software (and blessed or directly maintained by the 'upstream,' in this case NVIDIA), which means that this may be out of scope for the project. In that case, you would be best off partnering with NVIDIA to create more suitable base images/improve their images, as DOI/docker-library is the only Docker Inc. sponsored image-authoring project/process. |
Beta Was this translation helpful? Give feedback.
-
Description
It would be great to have light weight NVIDIA docker images.
Example:
To support state of the AI libraries, the developer image may be needed.
Wondering if there is a common request yet from docker community - To have a light weight/slim version of these images?
With the explosion of generative AI use-cases, I wonder if it makes sense for the community to think about this. This could have a big impact on cost savings
In my research, I have seen docker enthusiasts trying to build a solution using slim as the base image, but I couldn't come across a good solution.
Wanted to get your thoughts on this request
Beta Was this translation helpful? Give feedback.
All reactions