mlflow--mlflow/tests/pyfunc/test_pyfunc_model_config.py

import os

import pytest
import yaml

import mlflow
from mlflow.models import Model


@pytest.fixture
def model_path(tmp_path):
    return os.path.join(tmp_path, "model")


@pytest.fixture
def model_config():
    return {
        "use_gpu": True,
        "temperature": 0.9,
        "timeout": 300,
    }


def _load_pyfunc(path):
    return TestModel()


class TestModel(mlflow.pyfunc.PythonModel):
    def predict(self, context, model_input, params=None):
        return model_input


class InferenceContextModel(mlflow.pyfunc.PythonModel):
    def predict(self, context, model_input, params=None):
        # This mock class returns the internal inference configuration keys and values available
        return context.model_config.items()


def test_save_with_model_config(model_path, model_config):
    model = InferenceContextModel()
    mlflow.pyfunc.save_model(model_path, python_model=model, model_config=model_config)

    loaded_model = mlflow.pyfunc.load_model(model_uri=model_path)

    assert loaded_model.model_config
    assert set(model_config.keys()) == set(loaded_model.model_config)
    assert all(loaded_model.model_config[k] == v for k, v in model_config.items())
    assert all(loaded_model.model_config[k] == v for k, v in loaded_model.predict([[0]]))


@pytest.mark.parametrize(
    "model_config_path",
    [
        os.path.abspath("tests/pyfunc/sample_code/config.yml"),
        "tests/pyfunc/../pyfunc/sample_code/config.yml",
    ],
)
def test_save_with_model_config_path(model_path, model_config, model_config_path):
    model = InferenceContextModel()
    mlflow.pyfunc.save_model(model_path, python_model=model, model_config=model_config_path)

    loaded_model = mlflow.pyfunc.load_model(model_uri=model_path)

    assert loaded_model.model_config
    assert set(model_config.keys()) == set(loaded_model.model_config)
    assert all(loaded_model.model_config[k] == v for k, v in model_config.items())
    assert all(loaded_model.model_config[k] == v for k, v in loaded_model.predict([[0]]))


def test_override_model_config(model_path, model_config):
    model = TestModel()
    inference_override = {"timeout": 400}

    mlflow.pyfunc.save_model(model_path, python_model=model, model_config=model_config)
    loaded_model = mlflow.pyfunc.load_model(model_uri=model_path, model_config=inference_override)

    assert loaded_model.model_config["timeout"] == 400
    assert all(loaded_model.model_config[k] == v for k, v in inference_override.items())


@pytest.mark.parametrize(
    "model_config_path",
    [
        os.path.abspath("tests/pyfunc/sample_code/config.yml"),
        "tests/pyfunc/../pyfunc/sample_code/config.yml",
    ],
)
def test_override_model_config_path(tmp_path, model_path, model_config_path):
    model = TestModel()
    inference_override = {"timeout": 400}
    config_path = tmp_path / "config.yml"
    config_path.write_text(yaml.dump(inference_override))

    mlflow.pyfunc.save_model(model_path, python_model=model, model_config=model_config_path)
    loaded_model = mlflow.pyfunc.load_model(model_uri=model_path, model_config=str(config_path))

    assert loaded_model.model_config["timeout"] == 400
    assert all(loaded_model.model_config[k] == v for k, v in inference_override.items())


def test_override_model_config_ignore_invalid(model_path, model_config):
    model = TestModel()
    inference_override = {"invalid_key": 400}

    mlflow.pyfunc.save_model(model_path, python_model=model, model_config=model_config)
    loaded_model = mlflow.pyfunc.load_model(model_uri=model_path, model_config=inference_override)

    assert loaded_model.predict([[5]])
    assert all(k not in loaded_model.model_config for k in inference_override.keys())


@pytest.mark.parametrize(
    "model_config_path",
    [
        os.path.abspath("tests/pyfunc/sample_code/config.yml"),
        "tests/pyfunc/../pyfunc/sample_code/config.yml",
    ],
)
def test_override_model_config_path_ignore_invalid(tmp_path, model_path, model_config_path):
    model = TestModel()
    inference_override = {"invalid_key": 400}
    config_path = tmp_path / "config.yml"
    config_path.write_text(yaml.dump(inference_override))

    mlflow.pyfunc.save_model(model_path, python_model=model, model_config=model_config_path)
    loaded_model = mlflow.pyfunc.load_model(model_uri=model_path, model_config=str(config_path))

    assert loaded_model.predict([[5]])
    assert all(k not in loaded_model.model_config for k in inference_override.keys())


def test_pyfunc_without_model_config(model_path, model_config):
    model = TestModel()
    mlflow.pyfunc.save_model(model_path, python_model=model)

    loaded_model = mlflow.pyfunc.load_model(model_uri=model_path, model_config=model_config)

    assert loaded_model.predict([[5]])
    assert not loaded_model.model_config


def test_pyfunc_loader_without_model_config(model_path):
    mlflow.pyfunc.save_model(
        path=model_path,
        data_path=".",
        loader_module=__name__,
        code_paths=[__file__],
        mlflow_model=Model(run_id="test", artifact_path="testtest"),
    )

    inference_override = {"invalid_key": 400}
    pyfunc_model = mlflow.pyfunc.load_model(model_path, model_config=inference_override)

    assert not pyfunc_model.model_config