-
Notifications
You must be signed in to change notification settings - Fork 36
Update RAPIDS containers to Ubuntu 24.04 #534
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Conversation
Will also need to update the devcontainer.json files here to remove the |
@trxcllnt Do you know what would cause this error? I'm not sure what to do.
|
That tag doesn't exist yet. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Recommended changes
Looks like there aren't any CUDA 12.0 Ubuntu24.04 apt repos, so probably gotta stick with 22.04 for those (if we still need them). |
RAPIDS doesn't need to build with CUDA 12.0. We are okay with requiring 12.9+ to build. |
Co-authored-by: Paul Taylor <[email protected]>
Co-authored-by: Paul Taylor <[email protected]>
There's an error coming up, probably from rapidsai/rmm#1980. I think this would point to an issue in CMake code somewhere, like missing
|
It's installing |
The problem is coming from Do we need to make the devcontainers aware of xgboost, or add some special exclusion for librmm? |
conda doesn't let us exclude libraries installed by our dependencies. We either have to add xgboost as a library we build from source, or exclude xgboost from cuML's dependencies (ideally in |
Great. I filed a PR here with that change: rapidsai/cuml#7006 |
Unified conda devcontainers use the `all` file key to generate their dependency lists for each repository. Currently cuml includes `xgboost` in its `all` file key via the `test_python` dependency list. Unfortunately, this pulls in a conda dependency on `librmm` via `libxgboost`. This is problematic because we must build `librmm` from source in our unified devcontainers. It seems like `xgboost` is a soft test dependency. In this PR, I add a new dependency file key called `devcontainers` that excludes `xgboost`. Developer conda environments, CI workflows, and pip devcontainers are unaffected by this change. Conda devcontainers will now require the user to manually install `xgboost` in order to run tests that need it, and this is noted in `BUILD.md`. This is needed to solve the problems discussed in rapidsai/devcontainers#534 (comment). Once this is merged, I will push a change to rapidsai/devcontainers#536 to use the new `devcontainers` file key. Authors: - Bradley Dice (https://github.com/bdice) Approvers: - Paul Taylor (https://github.com/trxcllnt) - James Lamb (https://github.com/jameslamb) - Simon Adorf (https://github.com/csadorf) URL: #7006
@trxcllnt Can you re-review? |
Currently nightly builds fail for cuDF because of the use of
<format>
. This requires GCC 13 or newer.The devcontainers are getting the OS default version of GCC (because RAPIDS devcontainers don't specify a version). The OS is Ubuntu 22.04 which ships with GCC 11. Updating to Ubuntu 24.04 should solve this problem. Alternatively we can pin the GCC version, but I like being able to use system-default toolchains where possible.