def test_content_types(tensor_spec: TensorSpec, request_input: RequestInput):
input_schema = Schema(inputs=[tensor_spec])
inference_request = InferenceRequest(
parameters=Parameters(content_type=PandasCodec.ContentType),
inputs=[request_input],
)
data = decode_inference_request(inference_request)
# _enforce_schema will raise if something fails
_enforce_schema(data, input_schema)
async def test_predict_pytorch(runtime_pytorch: MLflowRuntime):
# The model used here is the MNIST pytorch example in mlflow:
# https://github.com/mlflow/mlflow/tree/master/examples/pytorch/MNIST
# input is a 28*28 image
data = np.random.randn(1, 28 * 28).astype(np.float32)
inference_request = InferenceRequest(
parameters=Parameters(content_type=NumpyCodec.ContentType),
inputs=[
RequestInput(
name="predict",
shape=data.shape,
data=data.tolist(),
datatype="FP32",
)
],
)
response = await runtime_pytorch.predict(inference_request)
outputs = response.outputs
assert len(outputs) == 1
assert outputs[0].name == DefaultOutputName
import pandas as pd
from typing import Any
from mlserver.types import RequestInput, Parameters, InferenceRequest
from mlserver.codecs import NumpyCodec, StringCodec, PandasCodec
from mlserver.codecs.middleware import DecodedParameterName, codec_middleware
from mlserver.settings import ModelSettings
@pytest.mark.parametrize(
"inference_request, expected",
[
(
InferenceRequest(
parameters=Parameters(content_type=PandasCodec.ContentType),
inputs=[
RequestInput(
name="foo",
shape=[2, 2],
data=[1, 2, 3, 4],
datatype="INT32",
parameters=Parameters(
content_type=NumpyCodec.ContentType),
),
],
),
pd.DataFrame({"foo": [[1, 2], [3, 4]]}),
),
(
InferenceRequest(
from mlserver.types import RequestInput, Parameters, TensorData
from mlserver.codecs import NumpyCodec, StringCodec
from mlserver.model import MLModel
@pytest.mark.parametrize(
"request_input,expected",
[
(
RequestInput(
name="foo",
shape=[2, 2],
data=[1, 2, 3, 4],
datatype="INT32",
parameters=Parameters(
content_type=NumpyCodec.ContentType,
_decoded_payload=np.array([[1, 2], [3, 4]]),
),
),
np.array([[1, 2], [3, 4]]),
),
(
RequestInput(
name="foo",
shape=[17],
data=b"my unicode string",
datatype="BYTES",
parameters=Parameters(
content_type=StringCodec.ContentType,
_decoded_payload="my unicode string",
),
),
)
from mlserver.batching.requests import BatchedRequests
@pytest.mark.parametrize(
"request_inputs, expected_request_input, expected_minibatch_sizes",
[
(
{
"req-1":
RequestInput(
name="foo",
datatype="INT32",
shape=[1, 3],
data=[1, 2, 3],
parameters=Parameters(content_type="np"),
),
"req-2":
RequestInput(
name="foo",
datatype="INT32",
shape=[1, 3],
data=[4, 5, 6],
),
"req-3":
RequestInput(
name="foo",
datatype="INT32",
shape=[1, 3],
data=[7, 8, 9],
parameters=Parameters(foo="bar"),
)
def test_to_metadata_tensors(schema: Schema, expected: List[MetadataTensor]):
metadata_tensors = to_metadata_tensors(schema)
assert metadata_tensors == expected
@pytest.mark.parametrize(
"tensor_spec, request_input",
[
(
ColSpec(name="foo", type=DataType.boolean),
RequestInput(
name="foo",
datatype="BOOL",
parameters=Parameters(content_type=NumpyCodec.ContentType),
shape=[2],
data=[True, False],
),
),
(
ColSpec(name="foo", type=DataType.string),
RequestInput(
name="foo",
datatype="BYTES",
parameters=Parameters(content_type=StringCodec.ContentType),
shape=[2, 11],
data=b"hello worldhello world",
),
),
(
model_version=expected.model_version)
assert inference_response == expected
@pytest.mark.parametrize(
"inference_request, expected",
[
(
InferenceRequest(inputs=[
RequestInput(
name="a",
data=[1, 2, 3],
datatype="FP32",
shape=[1, 3],
parameters=Parameters(
_decoded_payload=np.array([[1, 2, 3]])),
),
RequestInput(
name="b",
data=b"hello world",
datatype="BYTES",
shape=[1, 11],
parameters=Parameters(_decoded_payload=["hello world"]),
),
]),
pd.DataFrame({
"a": [np.array([1, 2, 3])],
"b": ["hello world"]
}),
),
(
RequestInput(
name="foo", shape=[2, 2], data=[1, 2, 3, 4], datatype="INT32"
)
]
),
np.array([[1, 2], [3, 4]]),
),
(
InferenceRequest(
inputs=[
RequestInput(
name="foo",
shape=[2, 2],
data=[1, 2, 3, 4],
datatype="INT32",
parameters=Parameters(**{DecodedParameterName: np.array([23])}),
)
]
),
np.array([23]),
),
],
)
def test_first_input_decode(inference_request: InferenceRequest, expected: np.ndarray):
inference_request.inputs = [inference_request.inputs[0]]
first_input = NumpyRequestCodec.decode(inference_request)
np.testing.assert_equal(first_input, expected)
def test_first_input_error(inference_request: InferenceRequest):