async def test_name_fallback(model_folder: str, model_repository: ModelRepository):
# Create empty model-settings.json file
model_settings = ModelSettings()
model_settings_path = os.path.join(model_folder, DEFAULT_MODEL_SETTINGS_FILENAME)
with open(model_settings_path, "w") as model_settings_file:
d = model_settings.dict()
del d["name"]
d["implementation"] = get_import_path(d["implementation"])
json.dump(d, model_settings_file)
model_settings = model_repository._load_model_settings(model_settings_path)
assert model_settings.name == os.path.basename(model_folder)
def test_model_settings_from_env(monkeypatch):
model_name = "foo-model"
model_version = "v0.1.0"
model_uri = "/mnt/models/my-model"
monkeypatch.setenv("mlserver_model_name", model_name)
monkeypatch.setenv("mlserver_model_version", model_version)
monkeypatch.setenv("mlserver_model_uri", model_uri)
model_settings = ModelSettings()
model_settings.parameters = ModelParameters()
assert model_settings.name == model_name
assert model_settings.version == model_version
assert model_settings.parameters.uri == model_uri
def case_custom_model(custom_model: Model) -> ModelSettings:
save(custom_model, save_env=False)
model_uri = custom_model.details.local_folder
return ModelSettings(
name="custom-model",
parameters=ModelParameters(uri=model_uri),
)
def case_wrapped_class_instance(inference_pipeline_class) -> ModelSettings:
save(inference_pipeline_class, save_env=False)
model_uri = inference_pipeline_class.pipeline.details.local_folder
return ModelSettings(
name="wrapped-class-instance",
parameters=ModelParameters(uri=model_uri),
)
async def mlserver_runtime(model_settings: ModelSettings) -> InferenceRuntime:
if is_lazy(model_settings):
# NOTE: Some times pytest-cases may return a "LazyValue"
model_settings = model_settings.get(request_or_item=mlserver_runtime)
_runtime = InferenceRuntime(model_settings)
await _runtime.load()
return _runtime
def case_wrapped_class(inference_pipeline_class) -> ModelSettings:
MyClass = inference_pipeline_class.__class__
save(MyClass, save_env=False)
model_uri = MyClass.pipeline.details.local_folder
return ModelSettings(
name="wrapped-class",
parameters=ModelParameters(uri=model_uri),
)
def xgboost_model(xgboost_model_uri: str) -> XGBoostModel:
model_settings = ModelSettings(
name="xgboost-model",
version="v1.2.3",
parameters=ModelParameters(uri=xgboost_model_uri),
)
model = XGBoostModel(model_settings)
model.load()
return model
def sklearn_model(sklearn_model_uri: str) -> SKLearnModel:
model_settings = ModelSettings(
name="sklearn-model",
version="v1.2.3",
parameters=ModelParameters(uri=sklearn_model_uri),
)
model = SKLearnModel(model_settings)
model.load()
return model
async def test_ready(data_plane, model_registry, ready):
model_settings = ModelSettings(
name="sum-model-2", parameters=ModelParameters(version="v1.2.3"))
new_model = SumModel(model_settings)
await model_registry.load(new_model)
new_model.ready = ready
all_ready = await data_plane.ready()
assert all_ready == ready
def case_async_custom_model() -> ModelSettings:
@aio.model(name="async-custom-model", platform=ModelFramework.Custom)
async def _custom_model(payload: np.ndarray) -> np.ndarray:
return payload.sum(keepdims=True)
save(_custom_model, save_env=False)
model_uri = _custom_model.details.local_folder
return ModelSettings(
name="async-custom-model",
parameters=ModelParameters(uri=model_uri),
)
def model_settings_pytorch_fixed(pytorch_model_uri) -> ModelSettings:
return ModelSettings(
name="mlflow-model",
parameters=ModelParameters(uri=pytorch_model_uri),
)
def model_settings(model_uri: str) -> ModelSettings:
return ModelSettings(
name="mlflow-model",
parameters=ModelParameters(uri=model_uri),
)
def model_settings(model_uri: str) -> ModelSettings:
return ModelSettings(
name="xgboost-model",
parameters=ModelParameters(uri=model_uri, version="v1.2.3"),
)
def model_settings(pipeline_uri: str) -> ModelSettings:
return ModelSettings(
name="sum-pipeline",
parameters=ModelParameters(uri=pipeline_uri),
)
def xgboost_model_settings(xgboost_model_uri: str) -> ModelSettings:
return ModelSettings(
name="xgboost-model",
version="v1.2.3",
parameters=ModelParameters(uri=xgboost_model_uri),
)
def sklearn_model_settings(sklearn_model_uri: str) -> ModelSettings:
return ModelSettings(
name="sklearn-model",
version="v1.2.3",
parameters=ModelParameters(uri=sklearn_model_uri),
)