introduce libcuml wheels

[jameslamb]

Replaces #6006, contributes to rapidsai/build-planning#33.

Proposes packaging libcuml as a wheel, which is then re-used by cuml-cu{11,12} wheels.

Notes for Reviewers

Benefits of these changes

• smaller wheels (see "Size Changes" below)
• faster compile times
  • no more re-compiling RAFT, thanks to introduce libraft wheels raft#2531
• less use of CI resources (only compiling once per CPU architecture / CUDA versions, instead of once per those + Python minor version)
• other benefits mentioned in Support dynamic linking between RAPIDS wheels build-planning#33

Wheel contents

libcuml:

• libcuml++.so (shared library) and its headers
• libcumlprims_mg.so (shared library) and its headers
• other vendored dependencies (CCCL, fmt)

cuml:

• cuml Python / Cython code and compiled Cython extensions

Dependency Flows

In short.... libcuml contains libcuml.so and libcumlprims_mg.so dynamic libraries and the headers to link against them.

• Anything that needs to link against cuML at build time pulls in libcugraph wheels as a build dependency.
• Anything that needs cuML's symbols at runtime pulls it in as a runtime dependency, and calls libcuml.load_library().

For more details and some flowcharts, see rapidsai/build-planning#33 (comment)

Size changes (CUDA 12, Python 3.12, x86_64)

wheel	num files (before)	num files (this PR)	size (before)	size (this PR)
libcuml	---	1766	---	289M
cuml	442	441	527M	9M
TOTAL	442	2207	527M	298M

NOTES: size = compressed, "before" = 2025-01-22 nightlies

how I calculated those (click me)

docker run \
    --rm \
    --network host \
    --env RAPIDS_NIGHTLY_DATE=2025-01-22 \
    --env CUML_NIGHTLY_SHA=01e19bba9821954b062a04fbf31d3522afa4b0b1 \
    --env CUML_PR="pull-request/6199" \
    --env CUML_PR_SHA="9d5100ec4589e20230a31817518427efa1e49c6d" \
    --env RAPIDS_PY_CUDA_SUFFIX=cu12 \
    --env WHEEL_DIR_BEFORE=/tmp/wheels-before \
    --env WHEEL_DIR_AFTER=/tmp/wheels-after \
    -it rapidsai/ci-wheel:cuda12.5.1-rockylinux8-py3.12 \
    bash

# --- nightly wheels --- #
mkdir -p ./wheels-before

export RAPIDS_BUILD_TYPE=branch
export RAPIDS_REF_NAME="branch-25.02"

# cuml
RAPIDS_PY_WHEEL_NAME="cuml_${RAPIDS_PY_CUDA_SUFFIX}" \
RAPIDS_REPOSITORY=rapidsai/cuml \
RAPIDS_SHA=${CUML_NIGHTLY_SHA} \
    rapids-download-wheels-from-s3 python ./wheels-before

# --- wheels from CI --- #
mkdir -p ./wheels-after

export RAPIDS_BUILD_TYPE="pull-request"

# libcuml
RAPIDS_PY_WHEEL_NAME="libcuml_${RAPIDS_PY_CUDA_SUFFIX}" \
RAPIDS_REPOSITORY=rapidsai/cuml \
RAPIDS_REF_NAME="${CUML_PR}" \
RAPIDS_SHA="${CUML_PR_SHA}" \
    rapids-download-wheels-from-s3 cpp ./wheels-after

# cuml
RAPIDS_PY_WHEEL_NAME="cuml_${RAPIDS_PY_CUDA_SUFFIX}" \
RAPIDS_REPOSITORY=rapidsai/cuml \
RAPIDS_REF_NAME="${CUML_PR}" \
RAPIDS_SHA="${CUML_PR_SHA}" \
    rapids-download-wheels-from-s3 python ./wheels-after

pip install pydistcheck
pydistcheck \
    --inspect \
    --select 'distro-too-large-compressed' \
    ./wheels-before/*.whl \
| grep -E '^checking|files: | compressed' \
> ./before.txt

# get more exact sizes
du -sh ./wheels-before/*

pydistcheck \
    --inspect \
    --select 'distro-too-large-compressed' \
    ./wheels-after/*.whl \
| grep -E '^checking|files: | compressed' \
> ./after.txt

# get more exact sizes
du -sh ./wheels-after/*

How I tested this

These other PRs:

• add libcuml Python builds devcontainers#442

[jameslamb]

conda-python-tests-singlegpu has been running for a LONG time here... is it normal for that to happen sometimes?

On this PR, I see:

• 11.8.0, 3.10, amd64, rockylinux8, v100, earliest-driver, oldest-deps (still running - 4h37m)
• 12.5.1, 3.12, amd64, ubuntu24.04, v100, latest-driver, latest-deps (finished - 4h4m)
• 12.0.1, 3.11, arm64, ubuntu20.04, a100, latest-driver, latest-deps (finished - 56m22s)

Those times do not include time spent waiting for a runner to pick up the jobs... that's all time occupying a GPU runner.

I see that happening on other PRs too, for example this in-progress build from #6238: https://github.com/rapidsai/cuml/actions/runs/12931830455/job/36069533961

Not sure if #6247 is the root cause... that refers to a "hang" but doesn't have other details.

[jameslamb]

Put up #6251 tracking the long CI times, I see them even on whitespace-only PRs like #6238 so don't think they're caused by the changes in this PR.

[vyasr]

Sweet everything is finally passing here!

[jameslamb]

Sweet everything is finally passing here!

I think this is ready! But holding off on merging because @dantegd indicated he also wanted to review. (update: talked offline, this is good to merge)

[jameslamb]

/merge