How to check if XGBoost uses the GPU

I'm writing a pytest file to check if my machine learning libraries use the GPU. For Tensorflow I can check this with tf.config.list_physical_devices(). For XGBoost I've so far checked it by looking at GPU utilization (nvdidia-smi) while running my software. But how can I check this in a simple test? Something similar to the test I have for Tensorflow would do.

import pytest
import tensorflow as tf
import xgboost

# Marking all tests to be GPU dependent
pytestmark = pytest.mark.gpu

def test_tf_finds_gpu():
    """Check if Tensorflow finds the GPU."""
    assert tf.config.list_physical_devices("GPU")

def test_xgb_finds_gpu():
    """Check if XGBoost finds the GPU."""
    ...
    # What can I write here?

Solution

Note that tree_method="gpu_hist" is deprecated and will stop / has stopped working since xgboost==2.0.0. Histogram type and device are currently split into two parameters: tree_method (an unfortunate overwriting of the existing parameter, but with a different set of permitted levels) and a new one called device:

import numpy as np
import xgboost as xgb

xgb_model = xgb.XGBRegressor( # tree_method="gpu_hist" # deprecated
    tree_method="hist",
    device="cuda"
)

X = np.random.rand(50, 2)
y = np.random.randint(2, size=50)

xgb_model.fit(X, y)

xgb_model

Output when GPU access works correctly (no warnings):

XGBRegressor(base_score=None, booster=None, callbacks=None,
             colsample_bylevel=None, colsample_bynode=None,
             colsample_bytree=None, device='cuda',
             early_stopping_rounds=None,
             enable_categorical=False, eval_metric=None, feature_types=None,
             gamma=None, grow_policy=None, importance_type=None,
             interaction_constraints=None, learning_rate=None, max_bin=None,
             max_cat_threshold=None, max_cat_to_onehot=None,
             max_delta_step=None, max_depth=None, max_leaves=None,
             min_child_weight=None, missing=nan, monotone_constraints=None,
             multi_strategy=None, n_estimators=None, n_jobs=None,
             num_parallel_tree=None, random_state=None, ...)

vs.

No GPU access - a warning that the device argument has not been used:

[11:43:35] WARNING: ../src/learner.cc:767: 
Parameters: { "device" } are not used.