3.0.0 Release Candidate.

[trivialfis]

Hi everyone, we are delighted to announce a new release 3.0.0 for XGBoost in the upcoming weeks. We invite everyone to try out the release candidate (RC).

Draft of the release note: #11285

Feedback period: until the end of March 06, 2025. No new feature will be added to the release branch; only bug fixes will be backported. If you have specific items that you want to include in the upcoming release, please share them in the comment below. Let's ensure that:

• Each item has an associated ticket
• Blocking issue must be marked as blocking
• Breaking change must be marked as breaking

@dmlc/xgboost-committer

Available packages:

• Python package

pip install xgboost==3.0.0rc1

• JVM packages

Show instructions for CPU (Maven/SBT)

Maven

<dependencies>
  ...
  <dependency>
      <groupId>ml.dmlc</groupId>
      <artifactId>xgboost4j-spark_2.12</artifactId>
      <version>3.0.0-RC1</version>
  </dependency>
</dependencies>

<repositories>
  <repository>
    <id>XGBoost4J Release Repo</id>
    <name>XGBoost4J Release Repo</name>
    <url>https://s3-us-west-2.amazonaws.com/xgboost-maven-repo/release/</url>
  </repository>
</repositories>

SBT

libraryDependencies ++= Seq(
  "org.apache.spark" %% "spark-core" % "3.5.0",
  "org.apache.spark" %% "spark-sql" % "3.5.0",
  "org.apache.spark" %% "spark-mllib" % "3.5.0",
  "ml.dmlc" %% "xgboost4j-spark" % "3.0.0-RC1"
)

resolvers += ("XGBoost4J Release Repo"
  at "https://s3-us-west-2.amazonaws.com/xgboost-maven-repo/release/")

Show instructions for GPU (Maven/SBT)

Maven

<dependencies>
  ...
  <dependency>
      <groupId>ml.dmlc</groupId>
      <artifactId>xgboost4j-spark-gpu_2.12</artifactId>
      <version>3.0.0-RC1</version>
  </dependency>
</dependencies>

<repositories>
  <repository>
    <id>XGBoost4J Release Repo</id>
    <name>XGBoost4J Release Repo</name>
    <url>https://s3-us-west-2.amazonaws.com/xgboost-maven-repo/release/</url>
  </repository>
</repositories>

SBT

libraryDependencies ++= Seq(
  "org.apache.spark" %% "spark-core" % "3.5.0",
  "org.apache.spark" %% "spark-sql" % "3.5.0",
  "org.apache.spark" %% "spark-mllib" % "3.5.0",
  "ml.dmlc" %% "xgboost4j-spark-gpu" % "3.0.0-RC1"
)

resolvers += ("XGBoost4J Release Repo"
  at "https://s3-us-west-2.amazonaws.com/xgboost-maven-repo/release/")

• R package
  xgboost_r_gpu_linux_3.0.0.tar.gz: Download

1b2991672f34d642776ece69a7501b47405c77b4942084f88dc46801af46f23a  xgboost_r_gpu_linux.tar.gz

Backports

• Update release scripts #11289
• [dask] Fix with the latest Dask. #11291
• [pyspark] Support columnar input for cpu pipeline #11299
• Normalize the sampling-based ltr with num pairs instead of grad. #11322
• [EM] Disable the on_host option for CPU inputs. #11333
• Sycl. Fix InitEstimations for sycl-devices #11331

[terrytangyuan]

Great to see this! Thank you for your work on this.

[trivialfis]

@hcho3 Is it expected that the PyPI tag macosx_10_15_x86_64.macosx_11_0_x86_64.macosx_12_0_x86_64 is now simply macosx_10_15_x86_64?

xgboost_cpu seems missing for win64.

xgboost-cpu fails twine checks:

$ twine check /home/jiamingy/workspace/xgboost_dev/releases/dist/xgboost_cpu-3.0.0rc1-py3-none-manylinux2014_x86_64*
Checking 
/home/jiamingy/workspace/xgboost_dev/releases/dist/xgboost_cpu-3.0.0rc1-py3-none-manylinux2014_x86_64.whl: FAILED
ERROR    `long_description` has syntax errors in markup and would not be rendered on PyPI.                               line 5: Error: Undefined substitution referenced: "PyPI version".

Checking /home/jiamingy/workspace/xgboost_dev/releases/dist/xgboost_cpu-3.0.0rc1-py3-none-manylinux2014_x86_64:
ERROR    InvalidDistribution: Unknown distribution format: 'xgboost_cpu-3.0.0rc1-py3-none-manylin

[trivialfis]

Let me defer the xgboost-cpu package for now until all issues are fixed.

[trivialfis]

Python packages for xgboost are uploaded to PyPI.

[trivialfis]

Updated JVM instructions.

[hcho3]

I need to make time to update the script for Maven Central release.

[trivialfis]

Delaying the release to next week for performance testing.

[trivialfis]

@hcho3 Do you think we should continue to use a CPU variant for sdist now that we have xgboost-cpu.

xgboost/dev/release-artifacts.py

Line 126 in d63d98b

with open("ops/patch/remove_nccl_dep.patch", "r") as f:

If so, I will update the script to use

xgboost/ops/script/pypi_variants.py

Line 27 in d63d98b

def make_pyproject(variant: str) -> None:

Otherwise, we can skip this patch. Alternatively, we can submit sdist for both packages.

Open a PR to update the script: #11337 .

[trivialfis]

Uploaded all packages for PyPI.

@hcho3 I will look into a DMLC R-Universe release next week. We need to host a new repo in the dmlc org. r-universe-org/help#578

[trivialfis]

I will delete the stable branch now, and renew it if necessary.

update: No need to renew it, RTD picked up 3.0.0 as the stable version.

[trivialfis]

@hcho3 Could you please help upload the JVM packages when you are available?