def __init__(
self,
name: str,
user_func: Callable[[Any], Any] = None,
local_folder: str = None,
uri: str = None,
platform: ModelFramework = None,
inputs: ModelDataType = None,
outputs: ModelDataType = None,
conda_env: str = None,
runtime: Runtime = None,
):
self._name = name
self._user_func = user_func
self.conda_env = conda_env
if uri is None:
uri = ""
local_folder = self._get_local_folder(local_folder)
inputs, outputs = self._get_args(inputs, outputs)
self.details = ModelDetails(
name=name,
local_folder=local_folder,
uri=uri,
platform=platform,
inputs=inputs,
outputs=outputs,
)
self.cls = None
self.runtime = runtime
def test_kubernetes_spec_pipeline():
details = ModelDetails(
name="inference-pipeline",
platform=ModelFramework.TempoPipeline,
uri="gs://seldon/tempo",
local_folder="",
inputs=ModelDataArgs(args=[]),
outputs=ModelDataArgs(args=[]),
)
options = KubernetesOptions(namespace="production", replicas=1)
k8s_object = KubernetesSpec(details, KFServingV2Protocol(), options)
container_spec = _V2ContainerFactory.get_container_spec(details)
container_env = [{
"name": name,
"value": value
} for name, value in container_spec["environment"].items()]
expected = {
"apiVersion": "machinelearning.seldon.io/v1",
"kind": "SeldonDeployment",
"metadata": {
"name": details.name,
"namespace": options.namespace
},
"spec": {
"protocol":
"kfserving",
"predictors": [{
"componentSpecs": [{
"spec": {
"containers": [{
"name": "classifier",
"image": container_spec["image"],
"env": container_env,
"args": [],
}]
}
}],
"graph": {
"modelUri": details.uri,
"name": "classifier",
"type": "MODEL",
"implementation": "TRITON_SERVER",
"serviceAccountName": "tempo-pipeline",
},
"name":
"default",
"replicas":
options.replicas,
}],
},
}
assert k8s_object.spec == expected
def test_tensorflow_spec():
md = ModelDetails(
name="test",
local_folder="",
uri="",
platform=ModelFramework.Tensorflow,
inputs=ModelDataArgs(args=[]),
outputs=ModelDataArgs(args=[]),
)
protocol = SeldonProtocol()
options = KubernetesOptions(namespace="production", replicas=1)
runtime_options = RuntimeOptions(k8s_options=options)
model_spec = ModelSpec(model_details=md,
protocol=protocol,
runtime_options=runtime_options)
spec = get_container_spec(model_spec)
assert "image" in spec
assert "command" in spec
def test_model_spec():
ms = ModelSpec(
model_details=ModelDetails(
name="test",
local_folder="",
uri="",
platform=ModelFramework.XGBoost,
inputs=ModelDataArgs(args=[ModelDataArg(ty=str)]),
outputs=ModelDataArgs(args=[]),
),
protocol=V2Protocol(),
runtime_options=KFServingOptions().local_options,
)
s = ms.json()
j = json.loads(s)
ms2 = ModelSpec(**j)
assert isinstance(ms2.protocol, V2Protocol)
assert ms2.model_details.inputs.args[0].ty == str
def __init__(
self,
name: str,
runtime: Runtime = None,
local_folder: str = None,
uri: str = None,
platform: ModelFramework = None,
inputs: ModelDataType = None,
outputs: ModelDataType = None,
model_func: Callable[[Any], Any] = None,
):
super().__init__(name, model_func, runtime.get_protocol(), inputs,
outputs)
self._model_func = model_func
self._runtime = runtime
self._details = ModelDetails(
name=name,
local_folder=local_folder,
uri=uri,
platform=platform,
inputs=inputs,
outputs=outputs,
)
def test_kubernetes_spec_pipeline():
details = ModelDetails(
name="inference-pipeline",
platform=ModelFramework.TempoPipeline,
uri="gs://seldon/tempo",
local_folder="",
inputs=ModelDataArgs(args=[]),
outputs=ModelDataArgs(args=[]),
)
options = KubernetesOptions(namespace="production", replicas=1)
protocol = KFServingV2Protocol()
runtime_options = RuntimeOptions(k8s_options=options)
model_spec = ModelSpec(model_details=details,
protocol=protocol,
runtime_options=runtime_options)
k8s_object = KubernetesSpec(model_spec)
expected = {
"apiVersion": "machinelearning.seldon.io/v1",
"kind": "SeldonDeployment",
"metadata": {
"annotations": {
"seldon.io/tempo-description":
"",
"seldon.io/tempo-model":
'{"model_details": '
'{"name": '
'"inference-pipeline", '
'"local_folder": "", '
'"uri": '
'"gs://seldon/tempo", '
'"platform": "tempo", '
'"inputs": {"args": '
'[]}, "outputs": '
'{"args": []}, '
'"description": ""}, '
'"protocol": '
'"tempo.kfserving.protocol.KFServingV2Protocol", '
'"runtime_options": '
'{"runtime": null, '
'"docker_options": '
'{"defaultRuntime": '
'"tempo.seldon.SeldonDockerRuntime"}, '
'"k8s_options": '
'{"replicas": 1, '
'"minReplicas": null, '
'"maxReplicas": null, '
'"authSecretName": '
"null, "
'"serviceAccountName": '
"null, "
'"defaultRuntime": '
'"tempo.seldon.SeldonKubernetesRuntime", '
'"namespace": '
'"production"}, '
'"ingress_options": '
'{"ingress": '
'"tempo.ingress.istio.IstioIngress", '
'"ssl": false, '
'"verify_ssl": true}}}',
},
"labels": {
"seldon.io/tempo": "true"
},
"name": "inference-pipeline",
"namespace": "production",
},
"spec": {
"protocol":
"kfserving",
"predictors": [{
"graph": {
"modelUri": details.uri,
"name": "inference-pipeline",
"type": "MODEL",
"implementation": "TEMPO_SERVER",
"serviceAccountName": "tempo-pipeline",
},
"name": "default",
"replicas": options.replicas,
}],
},
}
assert k8s_object.spec == expected