def skip_if_pylint_unavailable():
return pytest.mark.skipif(
not _is_importable("pylint"),
reason="pylint is required to run tests in this module")
mlflow.pytorch.log_model(model, artifact_path="model")
model_uri = mlflow.get_artifact_uri("model")
resp = pyfunc_serve_and_score_model(
model_uri,
data[0],
pyfunc_scoring_server.CONTENT_TYPE_JSON_SPLIT_ORIENTED,
extra_args=EXTRA_PYFUNC_SERVING_TEST_ARGS,
)
scores = pd.DataFrame(json.loads(resp.content))
np.testing.assert_array_almost_equal(scores.values[:, 0],
_predict(model=model, data=data))
@pytest.mark.large
@pytest.mark.skipif(not _is_importable("transformers"),
reason="This test requires transformers")
def test_pyfunc_serve_and_score_transformers():
from transformers import BertModel, BertConfig # pylint: disable=import-error
class MyBertModel(BertModel):
def forward(self, *args, **kwargs): # pylint: disable=arguments-differ
return super().forward(*args, **kwargs).last_hidden_state
model = MyBertModel(
BertConfig(
vocab_size=16,
hidden_size=2,
num_hidden_layers=2,
num_attention_heads=2,
intermediate_size=2,
with mlflow.start_run():
mlflow.pytorch.log_model(model, artifact_path="model")
model_uri = mlflow.get_artifact_uri("model")
resp = pyfunc_serve_and_score_model(
model_uri,
data[0],
pyfunc_scoring_server.CONTENT_TYPE_JSON_SPLIT_ORIENTED,
extra_args=EXTRA_PYFUNC_SERVING_TEST_ARGS,
)
scores = pd.DataFrame(json.loads(resp.content))
np.testing.assert_array_almost_equal(scores.values[:, 0], _predict(model=model, data=data))
@pytest.mark.large
@pytest.mark.skipif(not _is_importable("transformers"), reason="This test requires transformers")
def test_pyfunc_serve_and_score_transformers():
from transformers import BertModel, BertConfig # pylint: disable=import-error
class MyBertModel(BertModel):
def forward(self, *args, **kwargs): # pylint: disable=arguments-differ
return super().forward(*args, **kwargs).last_hidden_state
model = MyBertModel(
BertConfig(
vocab_size=16,
hidden_size=2,
num_hidden_layers=2,
num_attention_heads=2,
intermediate_size=2,
)
"""Ensures that keras models without save_format can still be loaded."""
mlflow.keras.save_model(tf_keras_model, model_path, save_format="h5")
model_conf_path = os.path.join(model_path, "MLmodel")
model_conf = Model.load(model_conf_path)
flavor_conf = model_conf.flavors.get(mlflow.keras.FLAVOR_NAME)
assert flavor_conf is not None
del flavor_conf["save_format"]
model_conf.save(model_conf_path)
model_loaded = mlflow.keras.load_model(model_path)
assert tf_keras_model.to_json() == model_loaded.to_json()
@pytest.mark.large
@pytest.mark.skipif(
not _is_importable("transformers"),
reason=
"This test requires transformers, which is incompatible with Keras < 2.3.0",
)
def test_pyfunc_serve_and_score_transformers():
from transformers import BertConfig, TFBertModel # pylint: disable=import-error
bert = TFBertModel(
BertConfig(
vocab_size=16,
hidden_size=2,
num_hidden_layers=2,
num_attention_heads=2,
intermediate_size=2,
))
dummy_inputs = bert.dummy_inputs["input_ids"].numpy()
import pytest
from tests.helper_functions import _is_importable
pytestmark = pytest.mark.skipif(
not _is_importable("pylint"),
reason="pylint is required to run tests in this module")
@pytest.fixture(scope="module")
def test_case():
# Ref: https://pylint.pycqa.org/en/latest/how_tos/custom_checkers.html#testing-a-checker
import pylint.testutils
from pylint_plugins import PytestRaisesWithoutMatch
class TestPytestRaisesWithoutMatch(pylint.testutils.CheckerTestCase):
CHECKER_CLASS = PytestRaisesWithoutMatch
test_case = TestPytestRaisesWithoutMatch()
test_case.setup_method()
return test_case
def create_message(msg_id, node):
import pylint.testutils
return pylint.testutils.Message(msg_id=msg_id, node=node)
def extract_node(code):
import astroid