Conversation
|
spack/spack#43751 passed successfully so I think this image is good to go. There are still a few issues with spack/spack#39666 but those are due to aarch64, not due to the image. |
This reverts commit 41b0a81.
|
Not sure if there is something more robust, but usually I just build the image locally, then run the target stack install on it to make sure it passes, and call it good. I there are any changes to the spack.yaml that are needed, like pointing to the ROCm/Python externals, then you can make sure everything is working before pushing the image and it makes the Spack CI PR much easier. |
|
For now, the image I created here is good enough. It doesn't yet support external ROCm, but I'm told that AMD is working on supporting Spack-installed ROCm in the future. I'll leave ROCm to someone else. |
| rsync \ | ||
| unzip \ | ||
| wget \ | ||
| zlib1g-dev \ |
There was a problem hiding this comment.
Just to double check all of these are needed from the system.
autoconf
automake
gettext
libffi-deva
libssl-dev
libxml2-dev
m4
ncurses-dev
zlib1g-dev
3 participants
The idea is to create an image that can be used across all Linux ML pipelines (x86_64, aarch64, ppc64le) with a newer OS/GCC version. Some criteria:
/usr/bin/python3P.S. I don't know how to write or test a Dockerfile so if you have any suggestions, it's probably quicker to push to my branch than it is to explain it to me 😅