document changes in library-loading, update build-UCX-from-source docs#1099
document changes in library-loading, update build-UCX-from-source docs#1099rapids-bot[bot] merged 5 commits intorapidsai:branch-0.42from jameslamb:docs/library-loading
Conversation
pentschev
left a comment
There was a problem hiding this comment.
Thanks @jameslamb I've left a few comments, overall looks good.
| See https://docs.rapids.ai/install/ for the list of supported Python versions | ||
| for recent RAPIDS releases. |
There was a problem hiding this comment.
I think this is not the most user-friendly way to phrase this here, unfortunately we do not have UCX-Py explicitly documented in https://docs.rapids.ai/install/ which has a different versioning scheme than the rest of RAPIDS, and thus it would be difficult for someone to understand what Python version is supported for UCX-Py based solely on the RAPIDS version. I agree the way it was documented isn't optimal and I can't think of a better way to do this differently than what it was either.
There was a problem hiding this comment.
Sure, that makes sense. My main motivation here was to avoid a hard-coded list of versions, since duplicating that kind of information in multiple places makes it more likely that it'll become out of date.
That said.... we do have a rapids-reviser template set up to replace this line when we roll out add-a-new-Python-version updates to all of RAPIDS.
I'll revert this change based on your comments here.
There was a problem hiding this comment.
I agree that would be ideal, unfortunately I think it's not helpful for the users given the different branching.
How does the rapids-reviser work? Can we easily make it replace arbitrary code/docs lines?
There was a problem hiding this comment.
Alright, no prob! I've reverted that change here.
Can we easily make it replace arbitrary code/docs lines?
Yes totally! You give it a shell script and say "run this on these repos". This type of case is easy to automatically update there with sed, not a problem.
Co-authored-by: Peter Andreas Entschev <[email protected]>
|
Thanks @pentschev for continuing to work with me and @vyasr on these patterns for how these libraries find UCX 😁 |
|
/merge |
Contributes to rapidsai/build-planning#118
Caused by rapidsai/ucx-wheels#13
I originally came here to document the implications of rapidsai/ucx-wheels#13 in the docs, namely:
libucx-cu{11,12}wheel installed, then by defaultucx-pywill use UCX libraries from that wheelRAPIDS_LIBUCX_PREFER_SYTEM_LIBRARY=truecan be set to opt out of this and use a system installation insteadWhile doing that, I noticed some other opportunities for improvement in the installation docs:
Notes for Reviewers
How I tested this
Followed these instructions in a Docker container running on an x86_64 machine with 8 V100s.
docker run \ --rm \ --gpus 0 \ -v $(pwd):/opt/work \ -w /opt/work \ -it rapidsai/ci-conda:latest \ bashUsed
condato set up the build environment:conda create -n ucx -c conda-forge \ automake make libtool pkg-config \ "python=3.12" "setuptools>=64.0" "cython>=3.0.0" \ cuda-nvcc \ cuda-cudart-dev \ cuda-nvml-dev \ cuda-nvtx-dev \ cuda-version=12.5Ran variations of this code snippet to test my install:
python -c "import ucp; print(ucp.get_ucx_version())"