---
id: xgboost-ray
version: "0.1.19"
license: Apache 2.0
license_treatment: permissive
maintenance: abandoned
---

# xgboost-ray — A Ray backend for distributed XGBoost

License: permissive · Maintenance: abandoned · Downloads: 462.2K/mo

## What it is and what it does

XGBoost-Ray wraps XGBoost's training and prediction functions to run them in parallel across a Ray cluster. Instead of the standard xgb.DMatrix, you pass data via RayDMatrix, which shards it across Ray's object store, and configure distributed training with RayParams. It provides both a functional API (train/predict) and scikit-learn-compatible estimators, making it a drop-in replacement for single-machine XGBoost when you need to scale to multiple nodes or GPUs.

The package integrates with Ray Tune for distributed hyperparameter optimization and supports various data formats (Pandas DataFrames, NumPy arrays, CSV, Parquet). However, the project is archived and has not been updated since June 2024, meaning it may not work with recent versions of xgboost, ray, or Python.

Use it for:

- Train gradient boosting models on datasets too large for a single machine by distributing training across a Ray cluster.
- Run distributed hyperparameter tuning with Ray Tune, testing multiple configurations in parallel, each parallelized internally.
- Perform inference on large datasets by sharding them across Ray actors and predicting in parallel.
- Leverage multi-GPU training for XGBoost workloads on clusters with GPU nodes.
- Use RayDMatrix to load and shard Parquet files and other data formats across Ray's object store for distributed training.

## Worth the install?

AI-flagged interpretation of the facts above — verify before relying.

Distributes XGBoost training and inference across multiple nodes and GPUs using Ray as the underlying compute framework.

No. The package is archived and abandoned as of June 2024, with no maintenance for over a year. While it has low install friction and permissive licensing, the lack of updates means it will likely fail with current versions of xgboost, ray, and modern Python releases. Use only if you are locked into an older, pinned environment. For new projects, consider Ray's native XGBoost integrations or maintained alternatives.

## Install

pip install xgboost-ray
uv add xgboost-ray
poetry add xgboost-ray

## Installing xgboost-ray

Before you install: Low install friction with a pure Python wheel. However, the package is archived and abandoned as of June 2024, with no releases for over a year. Depends on ray, xgboost, numpy, pandas, wrapt, and packaging—all widely maintained libraries.

License in practice: Licensed under Apache 2.0 (permissive), allowing commercial use, modification, and distribution with minimal restrictions.

Quickstart:

pip install xgboost_ray

from xgboost_ray import RayDMatrix, RayParams, train

train_set = RayDMatrix(train_x, train_y)
bst = train(
    {"objective": "binary:logistic"},
    train_set,
    ray_params=RayParams(num_actors=2))

Requires Ray to be initialized and running; package is archived and no longer maintained.

Verify before relying:

- Compatibility with XGBoost versions released after September 2023.
- Whether the package works with current Ray versions and modern Python releases.
- Status of multi-GPU training on current hardware and CUDA versions.
- Whether the scikit-learn API estimators (RayXGBClassifier, RayXGBRegressor, etc.) remain functional with current dependencies.

## Package facts

- License: Apache 2.0 (permissive)
- Python support: unspecified
- Install friction: low
- Maintenance: abandoned
- Downloads: 462.2K/month (top 15,000 on PyPI)
- Known vulnerabilities: none known

## Tags

distributed xgboost training, xgboost on ray, multi-node gradient boosting, parallel xgboost, ray distributed machine learning, xgboost hyperparameter tuning ray, gpu accelerated xgboost, distributed-ml, archived

[View on SkillFed](https://skillfed.io/packages/xgboost-ray) · [View on PyPI](https://pypi.org/project/xgboost-ray/)