diff --git a/tests/python_package_test/test_engine.py b/tests/python_package_test/test_engine.py
index 3047565555e8..e90a9e44ff36 100644
--- a/tests/python_package_test/test_engine.py
+++ b/tests/python_package_test/test_engine.py
@@ -309,7 +309,7 @@ def test_missing_value_handle_none():
     assert evals_result['valid_0']['auc'][-1] == pytest.approx(ret)
 
 
-@pytest.mark.parametrize('use_quantized_grad', [True, False])
+@pytest.mark.parametrize('use_quantized_grad', [pytest.param(True, marks=pytest.mark.skipif(getenv('TASK', '') == 'cuda', reason='Skip because quantized training with categorical features is not supported for cuda version')), False])
 def test_categorical_handle(use_quantized_grad):
     x = [0, 1, 2, 3, 4, 5, 6, 7]
     y = [0, 1, 0, 1, 0, 1, 0, 1]
@@ -351,7 +351,7 @@ def test_categorical_handle(use_quantized_grad):
     assert evals_result['valid_0']['auc'][-1] == pytest.approx(ret)
 
 
-@pytest.mark.parametrize('use_quantized_grad', [True, False])
+@pytest.mark.parametrize('use_quantized_grad', [pytest.param(True, marks=pytest.mark.skipif(getenv('TASK', '') == 'cuda', reason='Skip because quantized training with categorical features is not supported for cuda version')), False])
 def test_categorical_handle_na(use_quantized_grad):
     x = [0, np.nan, 0, np.nan, 0, np.nan]
     y = [0, 1, 0, 1, 0, 1]
@@ -393,7 +393,7 @@ def test_categorical_handle_na(use_quantized_grad):
     assert evals_result['valid_0']['auc'][-1] == pytest.approx(ret)
 
 
-@pytest.mark.parametrize('use_quantized_grad', [True, False])
+@pytest.mark.parametrize('use_quantized_grad', [pytest.param(True, marks=pytest.mark.skipif(getenv('TASK', '') == 'cuda', reason='Skip because quantized training with categorical features is not supported for cuda version')), False])
 def test_categorical_non_zero_inputs(use_quantized_grad):
     x = [1, 1, 1, 1, 1, 1, 2, 2]
     y = [1, 1, 1, 1, 1, 1, 0, 0]