[Feature]: Add XGBoost integration for MetaXuda

[Perinban]

Request

Add support for running or accelerating XGBoost workloads through MetaXuda on Apple Silicon.

Use case

XGBoost is widely used for gradient-boosted tree training and inference. An integration with MetaXuda could allow Python users to use Apple GPUs for XGBoost-related workloads without requiring an NVIDIA CUDA system.

This would be useful for local machine-learning development, benchmarking, and experimentation on Apple Silicon.

Proposed approach

No response

Type

Framework integration

Area

Runtime

Suggested priority

P2 - Normal

Contribution

• I am willing to submit a pull request for this request.

Checks

• I searched existing issues before opening this request.

[Perinban]

Work is in progress on XGBoost integration for MetaXuda.

The current target is to include the initial supported integration in MetaXuda 3.0. Progress and any compatibility limitations will be tracked in this issue.

[Perinban]

For tracking, the current experimental XGBoost integration can now complete a small GPU regression workflow through MetaXuda.

The verification covered:
Creating an XGBoost DMatrix
Running three boosting rounds with device=cuda:0
Using the GPU hist tree method
Producing a decreasing training RMSE
Generating finite, non-constant predictions

The below screenshot shows the verification script and successful output.

This is an initial integration test rather than a claim that every XGBoost workload is supported.