mlserver - npm Package Compare versions

Comparing version 1.7.1 to 1.7.2rc1

mlserver/batching/requests.py

@@ -21,2 +21,3 @@ from collections import defaultdict, OrderedDict
     parameters = defaultdict(list)
+    payload_parameters = {}
     if payload.parameters is not None:
@@ -23,0 +24,0 @@ payload_parameters = payload.parameters.model_dump()

mlserver/cli/main.py

@@ -6,3 +6,2 @@ """
 import click
-import asyncio
 
@@ -15,3 +14,3 @@ from functools import wraps
 from ..logging import logger, configure_logger
-from ..utils import install_uvloop_event_loop
+from ..utils import AsyncManager
 
@@ -22,7 +21,11 @@ from .build import generate_dockerfile, build_image, write_dockerfile
 
+CTX_ASYNC_MGR_KEY = "async_manager"
 
+
 def click_async(f):
     @wraps(f)
     def wrapper(*args, **kwargs):
-        return asyncio.run(f(*args, **kwargs))
+        ctx = click.get_current_context()
+        async_mgr = ctx.obj[CTX_ASYNC_MGR_KEY]
+        return async_mgr.run(f(*args, **kwargs))
 
@@ -34,7 +37,8 @@ return wrapper
 @click.version_option()
-def root():
[e-mail]_context
+def root(ctx):
     """
     Command-line interface to manage MLServer models.
     """
-    pass
+    ctx.ensure_object(dict)
 
@@ -272,4 +276,4 @@
     configure_logger()
-    install_uvloop_event_loop()
-    root()
+    async_mgr = AsyncManager()
+    root(obj={CTX_ASYNC_MGR_KEY: async_mgr})
 
@@ -276,0 +280,0 @@

mlserver/codecs/__init__.py

@@ -0,3 +1,4 @@
+from typing import TYPE_CHECKING
+
 from .numpy import NumpyCodec, NumpyRequestCodec
-from .pandas import PandasCodec
 from .string import StringCodec, StringRequestCodec
@@ -27,2 +28,5 @@ from .base64 import Base64Codec
 
+if TYPE_CHECKING:  # pragma: no cover - type checking only
+    from .pandas import PandasCodec  # noqa: F401
+
 __all__ = [
@@ -53,1 +57,20 @@ "CodecError",
 ]
+
+
+def __getattr__(name: str):  # pragma: no cover - lightweight lazy import
+    if name == "PandasCodec":
+        return _load_pandas_codec()
+
+    raise AttributeError(f"module 'mlserver.codecs' has no attribute {name!r}")
+
+
+def _load_pandas_codec():
+    try:
+        from .pandas import PandasCodec as _PandasCodec  # Local import to stay optional
+    except Exception as exc:  # pragma: no cover - propagate useful context
+        raise ImportError(
+            "PandasCodec requires the optional 'pandas' dependency"
+        ) from exc
+
+    globals()["PandasCodec"] = _PandasCodec
+    return _PandasCodec

mlserver/codecs/pandas.py

@@ -118,3 +118,3 @@ import pandas as pd
         use_bytes: bool = True,
-        **kwargs
+        **kwargs,
     ) -> InferenceResponse:
@@ -121,0 +121,0 @@ outputs = cls.encode_outputs(payload, use_bytes=use_bytes)

mlserver/grpc/dataplane_pb2.pyi

@@ -7,10 +7,17 @@ """
 import builtins
+import collections.abc
 import google.protobuf.descriptor
 import google.protobuf.internal.containers
 import google.protobuf.message
+import sys
 import typing
-import typing_extensions
 
-DESCRIPTOR: google.protobuf.descriptor.FileDescriptor = ...
+if sys.version_info >= (3, 10):
+    import typing as typing_extensions
+else:
+    import typing_extensions
 
+DESCRIPTOR: google.protobuf.descriptor.FileDescriptor
+
[e-mail]
 class ServerLiveRequest(google.protobuf.message.Message):
@@ -21,3 +28,4 @@ """
 
-    DESCRIPTOR: google.protobuf.descriptor.Descriptor = ...
+    DESCRIPTOR: google.protobuf.descriptor.Descriptor
+
     def __init__(
@@ -27,10 +35,11 @@ self,
 
-global___ServerLiveRequest = ServerLiveRequest
+Global___ServerLiveRequest: typing_extensions.TypeAlias = ServerLiveRequest
 
[e-mail]
 class ServerLiveResponse(google.protobuf.message.Message):
-    DESCRIPTOR: google.protobuf.descriptor.Descriptor = ...
+    DESCRIPTOR: google.protobuf.descriptor.Descriptor
+
     LIVE_FIELD_NUMBER: builtins.int
-    live: builtins.bool = ...
+    live: builtins.bool
     """True if the inference server is live, false if not live."""
-
     def __init__(
@@ -41,8 +50,7 @@ self,
     ) -> None: ...
-    def ClearField(
-        self, field_name: typing_extensions.Literal["live", b"live"]
-    ) -> None: ...
+    def ClearField(self, field_name: typing.Literal["live", b"live"]) -> None: ...
 
-global___ServerLiveResponse = ServerLiveResponse
+Global___ServerLiveResponse: typing_extensions.TypeAlias = ServerLiveResponse
 
[e-mail]
 class ServerReadyRequest(google.protobuf.message.Message):
@@ -53,3 +61,4 @@ """
 
-    DESCRIPTOR: google.protobuf.descriptor.Descriptor = ...
+    DESCRIPTOR: google.protobuf.descriptor.Descriptor
+
     def __init__(
@@ -59,10 +68,11 @@ self,
 
-global___ServerReadyRequest = ServerReadyRequest
+Global___ServerReadyRequest: typing_extensions.TypeAlias = ServerReadyRequest
 
[e-mail]
 class ServerReadyResponse(google.protobuf.message.Message):
-    DESCRIPTOR: google.protobuf.descriptor.Descriptor = ...
+    DESCRIPTOR: google.protobuf.descriptor.Descriptor
+
     READY_FIELD_NUMBER: builtins.int
-    ready: builtins.bool = ...
+    ready: builtins.bool
     """True if the inference server is ready, false if not ready."""
-
     def __init__(
@@ -73,8 +83,7 @@ self,
     ) -> None: ...
-    def ClearField(
-        self, field_name: typing_extensions.Literal["ready", b"ready"]
-    ) -> None: ...
+    def ClearField(self, field_name: typing.Literal["ready", b"ready"]) -> None: ...
 
-global___ServerReadyResponse = ServerReadyResponse
+Global___ServerReadyResponse: typing_extensions.TypeAlias = ServerReadyResponse
 
[e-mail]
 class ModelReadyRequest(google.protobuf.message.Message):
@@ -85,32 +94,31 @@ """
 
-    DESCRIPTOR: google.protobuf.descriptor.Descriptor = ...
+    DESCRIPTOR: google.protobuf.descriptor.Descriptor
+
     NAME_FIELD_NUMBER: builtins.int
     VERSION_FIELD_NUMBER: builtins.int
-    name: typing.Text = ...
+    name: builtins.str
     """The name of the model to check for readiness."""
-
-    version: typing.Text = ...
+    version: builtins.str
     """The version of the model to check for readiness. If not given the
     server will choose a version based on the model and internal policy.
     """
-
     def __init__(
         self,
         *,
-        name: typing.Text = ...,
-        version: typing.Text = ...,
+        name: builtins.str = ...,
+        version: builtins.str = ...,
     ) -> None: ...
     def ClearField(
-        self,
-        field_name: typing_extensions.Literal["name", b"name", "version", b"version"],
+        self, field_name: typing.Literal["name", b"name", "version", b"version"]
     ) -> None: ...
 
-global___ModelReadyRequest = ModelReadyRequest
+Global___ModelReadyRequest: typing_extensions.TypeAlias = ModelReadyRequest
 
[e-mail]
 class ModelReadyResponse(google.protobuf.message.Message):
-    DESCRIPTOR: google.protobuf.descriptor.Descriptor = ...
+    DESCRIPTOR: google.protobuf.descriptor.Descriptor
+
     READY_FIELD_NUMBER: builtins.int
-    ready: builtins.bool = ...
+    ready: builtins.bool
     """True if the model is ready, false if not ready."""
-
     def __init__(
@@ -121,8 +129,7 @@ self,
     ) -> None: ...
-    def ClearField(
-        self, field_name: typing_extensions.Literal["ready", b"ready"]
-    ) -> None: ...
+    def ClearField(self, field_name: typing.Literal["ready", b"ready"]) -> None: ...
 
-global___ModelReadyResponse = ModelReadyResponse
+Global___ModelReadyResponse: typing_extensions.TypeAlias = ModelReadyResponse
 
[e-mail]
 class ServerMetadataRequest(google.protobuf.message.Message):
@@ -133,3 +140,4 @@ """
 
-    DESCRIPTOR: google.protobuf.descriptor.Descriptor = ...
+    DESCRIPTOR: google.protobuf.descriptor.Descriptor
+
     def __init__(
@@ -139,21 +147,20 @@ self,
 
-global___ServerMetadataRequest = ServerMetadataRequest
+Global___ServerMetadataRequest: typing_extensions.TypeAlias = ServerMetadataRequest
 
[e-mail]
 class ServerMetadataResponse(google.protobuf.message.Message):
-    DESCRIPTOR: google.protobuf.descriptor.Descriptor = ...
+    DESCRIPTOR: google.protobuf.descriptor.Descriptor
+
     NAME_FIELD_NUMBER: builtins.int
     VERSION_FIELD_NUMBER: builtins.int
     EXTENSIONS_FIELD_NUMBER: builtins.int
-    name: typing.Text = ...
+    name: builtins.str
     """The server name."""
-
-    version: typing.Text = ...
+    version: builtins.str
     """The server version."""
-
     @property
     def extensions(
         self,
-    ) -> google.protobuf.internal.containers.RepeatedScalarFieldContainer[typing.Text]:
+    ) -> google.protobuf.internal.containers.RepeatedScalarFieldContainer[builtins.str]:
         """The extensions supported by the server."""
-        pass
 
@@ -163,9 +170,9 @@ def __init__(
         *,
-        name: typing.Text = ...,
-        version: typing.Text = ...,
-        extensions: typing.Optional[typing.Iterable[typing.Text]] = ...,
+        name: builtins.str = ...,
+        version: builtins.str = ...,
+        extensions: collections.abc.Iterable[builtins.str] | None = ...,
     ) -> None: ...
     def ClearField(
         self,
-        field_name: typing_extensions.Literal[
+        field_name: typing.Literal[
             "extensions", b"extensions", "name", b"name", "version", b"version"
@@ -175,4 +182,5 @@ ],
 
-global___ServerMetadataResponse = ServerMetadataResponse
+Global___ServerMetadataResponse: typing_extensions.TypeAlias = ServerMetadataResponse
 
[e-mail]
 class ModelMetadataRequest(google.protobuf.message.Message):
@@ -183,53 +191,54 @@ """
 
-    DESCRIPTOR: google.protobuf.descriptor.Descriptor = ...
+    DESCRIPTOR: google.protobuf.descriptor.Descriptor
+
     NAME_FIELD_NUMBER: builtins.int
     VERSION_FIELD_NUMBER: builtins.int
-    name: typing.Text = ...
+    name: builtins.str
     """The name of the model."""
-
-    version: typing.Text = ...
+    version: builtins.str
     """The version of the model to check for readiness. If not given the
     server will choose a version based on the model and internal policy.
     """
-
     def __init__(
         self,
         *,
-        name: typing.Text = ...,
-        version: typing.Text = ...,
+        name: builtins.str = ...,
+        version: builtins.str = ...,
     ) -> None: ...
     def ClearField(
-        self,
-        field_name: typing_extensions.Literal["name", b"name", "version", b"version"],
+        self, field_name: typing.Literal["name", b"name", "version", b"version"]
     ) -> None: ...
 
-global___ModelMetadataRequest = ModelMetadataRequest
+Global___ModelMetadataRequest: typing_extensions.TypeAlias = ModelMetadataRequest
 
[e-mail]
 class ModelMetadataResponse(google.protobuf.message.Message):
-    DESCRIPTOR: google.protobuf.descriptor.Descriptor = ...
+    DESCRIPTOR: google.protobuf.descriptor.Descriptor
 
+    @typing.final
     class TensorMetadata(google.protobuf.message.Message):
         """Metadata for a tensor."""
 
-        DESCRIPTOR: google.protobuf.descriptor.Descriptor = ...
+        DESCRIPTOR: google.protobuf.descriptor.Descriptor
 
+        @typing.final
         class ParametersEntry(google.protobuf.message.Message):
-            DESCRIPTOR: google.protobuf.descriptor.Descriptor = ...
+            DESCRIPTOR: google.protobuf.descriptor.Descriptor
+
             KEY_FIELD_NUMBER: builtins.int
             VALUE_FIELD_NUMBER: builtins.int
-            key: typing.Text = ...
+            key: builtins.str
             @property
-            def value(self) -> global___InferParameter: ...
+            def value(self) -> Global___InferParameter: ...
             def __init__(
                 self,
                 *,
-                key: typing.Text = ...,
-                value: typing.Optional[global___InferParameter] = ...,
+                key: builtins.str = ...,
+                value: Global___InferParameter | None = ...,
             ) -> None: ...
             def HasField(
-                self, field_name: typing_extensions.Literal["value", b"value"]
+                self, field_name: typing.Literal["value", b"value"]
             ) -> builtins.bool: ...
             def ClearField(
-                self,
-                field_name: typing_extensions.Literal["key", b"key", "value", b"value"],
+                self, field_name: typing.Literal["key", b"key", "value", b"value"]
             ) -> None: ...
@@ -241,8 +250,6 @@
         PARAMETERS_FIELD_NUMBER: builtins.int
-        name: typing.Text = ...
+        name: builtins.str
         """The tensor name."""
-
-        datatype: typing.Text = ...
+        datatype: builtins.str
         """The tensor data type."""
-
         @property
@@ -257,3 +264,2 @@ def shape(
             """
-            pass
 
@@ -264,3 +270,3 @@ @property
         ) -> google.protobuf.internal.containers.MessageMap[
-            typing.Text, global___InferParameter
+            builtins.str, Global___InferParameter
         ]:
@@ -270,3 +276,2 @@ """Optional default parameters for input.
             """
-            pass
 
@@ -276,12 +281,12 @@ def __init__(
             *,
-            name: typing.Text = ...,
-            datatype: typing.Text = ...,
-            shape: typing.Optional[typing.Iterable[builtins.int]] = ...,
-            parameters: typing.Optional[
-                typing.Mapping[typing.Text, global___InferParameter]
-            ] = ...,
+            name: builtins.str = ...,
+            datatype: builtins.str = ...,
+            shape: collections.abc.Iterable[builtins.int] | None = ...,
+            parameters: (
+                collections.abc.Mapping[builtins.str, Global___InferParameter] | None
+            ) = ...,
         ) -> None: ...
         def ClearField(
             self,
-            field_name: typing_extensions.Literal[
+            field_name: typing.Literal[
                 "datatype",
@@ -298,21 +303,22 @@ b"datatype",
 
+    @typing.final
     class ParametersEntry(google.protobuf.message.Message):
-        DESCRIPTOR: google.protobuf.descriptor.Descriptor = ...
+        DESCRIPTOR: google.protobuf.descriptor.Descriptor
+
         KEY_FIELD_NUMBER: builtins.int
         VALUE_FIELD_NUMBER: builtins.int
-        key: typing.Text = ...
+        key: builtins.str
         @property
-        def value(self) -> global___InferParameter: ...
+        def value(self) -> Global___InferParameter: ...
         def __init__(
             self,
             *,
-            key: typing.Text = ...,
-            value: typing.Optional[global___InferParameter] = ...,
+            key: builtins.str = ...,
+            value: Global___InferParameter | None = ...,
         ) -> None: ...
         def HasField(
-            self, field_name: typing_extensions.Literal["value", b"value"]
+            self, field_name: typing.Literal["value", b"value"]
         ) -> builtins.bool: ...
         def ClearField(
-            self,
-            field_name: typing_extensions.Literal["key", b"key", "value", b"value"],
+            self, field_name: typing.Literal["key", b"key", "value", b"value"]
         ) -> None: ...
@@ -326,13 +332,11 @@
     PARAMETERS_FIELD_NUMBER: builtins.int
-    name: typing.Text = ...
+    name: builtins.str
     """The model name."""
-
+    platform: builtins.str
+    """The model's platform. See Platforms."""
     @property
     def versions(
         self,
-    ) -> google.protobuf.internal.containers.RepeatedScalarFieldContainer[typing.Text]:
+    ) -> google.protobuf.internal.containers.RepeatedScalarFieldContainer[builtins.str]:
         """The versions of the model available on the server."""
-        pass
-    platform: typing.Text = ...
-    """The model's platform. See Platforms."""
 
@@ -343,6 +347,5 @@ @property
     ) -> google.protobuf.internal.containers.RepeatedCompositeFieldContainer[
-        global___ModelMetadataResponse.TensorMetadata
+        Global___ModelMetadataResponse.TensorMetadata
     ]:
         """The model's inputs."""
-        pass
 
@@ -353,6 +356,5 @@ @property
     ) -> google.protobuf.internal.containers.RepeatedCompositeFieldContainer[
-        global___ModelMetadataResponse.TensorMetadata
+        Global___ModelMetadataResponse.TensorMetadata
     ]:
         """The model's outputs."""
-        pass
 
@@ -363,3 +365,3 @@ @property
     ) -> google.protobuf.internal.containers.MessageMap[
-        typing.Text, global___InferParameter
+        builtins.str, Global___InferParameter
     ]:
@@ -369,3 +371,2 @@ """Optional default parameters for the request / response.
         """
-        pass
 
@@ -375,18 +376,20 @@ def __init__(
         *,
-        name: typing.Text = ...,
-        versions: typing.Optional[typing.Iterable[typing.Text]] = ...,
-        platform: typing.Text = ...,
-        inputs: typing.Optional[
-            typing.Iterable[global___ModelMetadataResponse.TensorMetadata]
-        ] = ...,
-        outputs: typing.Optional[
-            typing.Iterable[global___ModelMetadataResponse.TensorMetadata]
-        ] = ...,
-        parameters: typing.Optional[
-            typing.Mapping[typing.Text, global___InferParameter]
-        ] = ...,
+        name: builtins.str = ...,
+        versions: collections.abc.Iterable[builtins.str] | None = ...,
+        platform: builtins.str = ...,
+        inputs: (
+            collections.abc.Iterable[Global___ModelMetadataResponse.TensorMetadata]
+            | None
+        ) = ...,
+        outputs: (
+            collections.abc.Iterable[Global___ModelMetadataResponse.TensorMetadata]
+            | None
+        ) = ...,
+        parameters: (
+            collections.abc.Mapping[builtins.str, Global___InferParameter] | None
+        ) = ...,
     ) -> None: ...
     def ClearField(
         self,
-        field_name: typing_extensions.Literal[
+        field_name: typing.Literal[
             "inputs",
@@ -407,4 +410,5 @@ b"inputs",
 
-global___ModelMetadataResponse = ModelMetadataResponse
+Global___ModelMetadataResponse: typing_extensions.TypeAlias = ModelMetadataResponse
 
[e-mail]
 class ModelInferRequest(google.protobuf.message.Message):
@@ -415,28 +419,30 @@ """
 
-    DESCRIPTOR: google.protobuf.descriptor.Descriptor = ...
+    DESCRIPTOR: google.protobuf.descriptor.Descriptor
 
+    @typing.final
     class InferInputTensor(google.protobuf.message.Message):
         """An input tensor for an inference request."""
 
-        DESCRIPTOR: google.protobuf.descriptor.Descriptor = ...
+        DESCRIPTOR: google.protobuf.descriptor.Descriptor
 
+        @typing.final
         class ParametersEntry(google.protobuf.message.Message):
-            DESCRIPTOR: google.protobuf.descriptor.Descriptor = ...
+            DESCRIPTOR: google.protobuf.descriptor.Descriptor
+
             KEY_FIELD_NUMBER: builtins.int
             VALUE_FIELD_NUMBER: builtins.int
-            key: typing.Text = ...
+            key: builtins.str
             @property
-            def value(self) -> global___InferParameter: ...
+            def value(self) -> Global___InferParameter: ...
             def __init__(
                 self,
                 *,
-                key: typing.Text = ...,
-                value: typing.Optional[global___InferParameter] = ...,
+                key: builtins.str = ...,
+                value: Global___InferParameter | None = ...,
             ) -> None: ...
             def HasField(
-                self, field_name: typing_extensions.Literal["value", b"value"]
+                self, field_name: typing.Literal["value", b"value"]
             ) -> builtins.bool: ...
             def ClearField(
-                self,
-                field_name: typing_extensions.Literal["key", b"key", "value", b"value"],
+                self, field_name: typing.Literal["key", b"key", "value", b"value"]
             ) -> None: ...
@@ -449,8 +455,6 @@
         CONTENTS_FIELD_NUMBER: builtins.int
-        name: typing.Text = ...
+        name: builtins.str
         """The tensor name."""
-
-        datatype: typing.Text = ...
+        datatype: builtins.str
         """The tensor data type."""
-
         @property
@@ -463,3 +467,2 @@ def shape(
             """The tensor shape."""
-            pass
 
@@ -470,9 +473,8 @@ @property
         ) -> google.protobuf.internal.containers.MessageMap[
-            typing.Text, global___InferParameter
+            builtins.str, Global___InferParameter
         ]:
             """Optional inference input tensor parameters."""
-            pass
 
         @property
-        def contents(self) -> global___InferTensorContents:
+        def contents(self) -> Global___InferTensorContents:
             """The input tensor data.
@@ -482,3 +484,2 @@ This field must not be specified if tensor contents are being specified
             """
-            pass
 
@@ -488,16 +489,16 @@ def __init__(
             *,
-            name: typing.Text = ...,
-            datatype: typing.Text = ...,
-            shape: typing.Optional[typing.Iterable[builtins.int]] = ...,
-            parameters: typing.Optional[
-                typing.Mapping[typing.Text, global___InferParameter]
-            ] = ...,
-            contents: typing.Optional[global___InferTensorContents] = ...,
+            name: builtins.str = ...,
+            datatype: builtins.str = ...,
+            shape: collections.abc.Iterable[builtins.int] | None = ...,
+            parameters: (
+                collections.abc.Mapping[builtins.str, Global___InferParameter] | None
+            ) = ...,
+            contents: Global___InferTensorContents | None = ...,
         ) -> None: ...
         def HasField(
-            self, field_name: typing_extensions.Literal["contents", b"contents"]
+            self, field_name: typing.Literal["contents", b"contents"]
         ) -> builtins.bool: ...
         def ClearField(
             self,
-            field_name: typing_extensions.Literal[
+            field_name: typing.Literal[
                 "contents",
@@ -516,26 +517,28 @@ b"contents",
 
+    @typing.final
     class InferRequestedOutputTensor(google.protobuf.message.Message):
         """An output tensor requested for an inference request."""
 
-        DESCRIPTOR: google.protobuf.descriptor.Descriptor = ...
+        DESCRIPTOR: google.protobuf.descriptor.Descriptor
 
+        @typing.final
         class ParametersEntry(google.protobuf.message.Message):
-            DESCRIPTOR: google.protobuf.descriptor.Descriptor = ...
+            DESCRIPTOR: google.protobuf.descriptor.Descriptor
+
             KEY_FIELD_NUMBER: builtins.int
             VALUE_FIELD_NUMBER: builtins.int
-            key: typing.Text = ...
+            key: builtins.str
             @property
-            def value(self) -> global___InferParameter: ...
+            def value(self) -> Global___InferParameter: ...
             def __init__(
                 self,
                 *,
-                key: typing.Text = ...,
-                value: typing.Optional[global___InferParameter] = ...,
+                key: builtins.str = ...,
+                value: Global___InferParameter | None = ...,
             ) -> None: ...
             def HasField(
-                self, field_name: typing_extensions.Literal["value", b"value"]
+                self, field_name: typing.Literal["value", b"value"]
             ) -> builtins.bool: ...
             def ClearField(
-                self,
-                field_name: typing_extensions.Literal["key", b"key", "value", b"value"],
+                self, field_name: typing.Literal["key", b"key", "value", b"value"]
             ) -> None: ...
@@ -545,5 +548,4 @@
         PARAMETERS_FIELD_NUMBER: builtins.int
-        name: typing.Text = ...
+        name: builtins.str
         """The tensor name."""
-
         @property
@@ -553,6 +555,5 @@ def parameters(
         ) -> google.protobuf.internal.containers.MessageMap[
-            typing.Text, global___InferParameter
+            builtins.str, Global___InferParameter
         ]:
             """Optional requested output tensor parameters."""
-            pass
 
@@ -562,33 +563,32 @@ def __init__(
             *,
-            name: typing.Text = ...,
-            parameters: typing.Optional[
-                typing.Mapping[typing.Text, global___InferParameter]
-            ] = ...,
+            name: builtins.str = ...,
+            parameters: (
+                collections.abc.Mapping[builtins.str, Global___InferParameter] | None
+            ) = ...,
         ) -> None: ...
         def ClearField(
             self,
-            field_name: typing_extensions.Literal[
-                "name", b"name", "parameters", b"parameters"
-            ],
+            field_name: typing.Literal["name", b"name", "parameters", b"parameters"],
         ) -> None: ...
 
+    @typing.final
     class ParametersEntry(google.protobuf.message.Message):
-        DESCRIPTOR: google.protobuf.descriptor.Descriptor = ...
+        DESCRIPTOR: google.protobuf.descriptor.Descriptor
+
         KEY_FIELD_NUMBER: builtins.int
         VALUE_FIELD_NUMBER: builtins.int
-        key: typing.Text = ...
+        key: builtins.str
         @property
-        def value(self) -> global___InferParameter: ...
+        def value(self) -> Global___InferParameter: ...
         def __init__(
             self,
             *,
-            key: typing.Text = ...,
-            value: typing.Optional[global___InferParameter] = ...,
+            key: builtins.str = ...,
+            value: Global___InferParameter | None = ...,
         ) -> None: ...
         def HasField(
-            self, field_name: typing_extensions.Literal["value", b"value"]
+            self, field_name: typing.Literal["value", b"value"]
         ) -> builtins.bool: ...
         def ClearField(
-            self,
-            field_name: typing_extensions.Literal["key", b"key", "value", b"value"],
+            self, field_name: typing.Literal["key", b"key", "value", b"value"]
         ) -> None: ...
@@ -603,15 +603,12 @@
     RAW_INPUT_CONTENTS_FIELD_NUMBER: builtins.int
-    model_name: typing.Text = ...
+    model_name: builtins.str
     """The name of the model to use for inferencing."""
-
-    model_version: typing.Text = ...
+    model_version: builtins.str
     """The version of the model to use for inference. If not given the
     server will choose a version based on the model and internal policy.
     """
-
-    id: typing.Text = ...
+    id: builtins.str
     """Optional identifier for the request. If specified will be
     returned in the response.
     """
-
     @property
@@ -621,6 +618,5 @@ def parameters(
     ) -> google.protobuf.internal.containers.MessageMap[
-        typing.Text, global___InferParameter
+        builtins.str, Global___InferParameter
     ]:
         """Optional inference parameters."""
-        pass
 
@@ -631,6 +627,5 @@ @property
     ) -> google.protobuf.internal.containers.RepeatedCompositeFieldContainer[
-        global___ModelInferRequest.InferInputTensor
+        Global___ModelInferRequest.InferInputTensor
     ]:
         """The input tensors for the inference."""
-        pass
 
@@ -641,3 +636,3 @@ @property
     ) -> google.protobuf.internal.containers.RepeatedCompositeFieldContainer[
-        global___ModelInferRequest.InferRequestedOutputTensor
+        Global___ModelInferRequest.InferRequestedOutputTensor
     ]:
@@ -647,3 +642,2 @@ """The requested output tensors for the inference. Optional, if not
         """
-        pass
 
@@ -674,3 +668,2 @@ @property
         """
-        pass
 
@@ -680,19 +673,22 @@ def __init__(
         *,
-        model_name: typing.Text = ...,
-        model_version: typing.Text = ...,
-        id: typing.Text = ...,
-        parameters: typing.Optional[
-            typing.Mapping[typing.Text, global___InferParameter]
-        ] = ...,
-        inputs: typing.Optional[
-            typing.Iterable[global___ModelInferRequest.InferInputTensor]
-        ] = ...,
-        outputs: typing.Optional[
-            typing.Iterable[global___ModelInferRequest.InferRequestedOutputTensor]
-        ] = ...,
-        raw_input_contents: typing.Optional[typing.Iterable[builtins.bytes]] = ...,
+        model_name: builtins.str = ...,
+        model_version: builtins.str = ...,
+        id: builtins.str = ...,
+        parameters: (
+            collections.abc.Mapping[builtins.str, Global___InferParameter] | None
+        ) = ...,
+        inputs: (
+            collections.abc.Iterable[Global___ModelInferRequest.InferInputTensor] | None
+        ) = ...,
+        outputs: (
+            collections.abc.Iterable[
+                Global___ModelInferRequest.InferRequestedOutputTensor
+            ]
+            | None
+        ) = ...,
+        raw_input_contents: collections.abc.Iterable[builtins.bytes] | None = ...,
     ) -> None: ...
     def ClearField(
         self,
-        field_name: typing_extensions.Literal[
+        field_name: typing.Literal[
             "id",
@@ -715,31 +711,34 @@ b"id",
 
-global___ModelInferRequest = ModelInferRequest
+Global___ModelInferRequest: typing_extensions.TypeAlias = ModelInferRequest
 
[e-mail]
 class ModelInferResponse(google.protobuf.message.Message):
-    DESCRIPTOR: google.protobuf.descriptor.Descriptor = ...
+    DESCRIPTOR: google.protobuf.descriptor.Descriptor
 
+    @typing.final
     class InferOutputTensor(google.protobuf.message.Message):
         """An output tensor returned for an inference request."""
 
-        DESCRIPTOR: google.protobuf.descriptor.Descriptor = ...
+        DESCRIPTOR: google.protobuf.descriptor.Descriptor
 
+        @typing.final
         class ParametersEntry(google.protobuf.message.Message):
-            DESCRIPTOR: google.protobuf.descriptor.Descriptor = ...
+            DESCRIPTOR: google.protobuf.descriptor.Descriptor
+
             KEY_FIELD_NUMBER: builtins.int
             VALUE_FIELD_NUMBER: builtins.int
-            key: typing.Text = ...
+            key: builtins.str
             @property
-            def value(self) -> global___InferParameter: ...
+            def value(self) -> Global___InferParameter: ...
             def __init__(
                 self,
                 *,
-                key: typing.Text = ...,
-                value: typing.Optional[global___InferParameter] = ...,
+                key: builtins.str = ...,
+                value: Global___InferParameter | None = ...,
             ) -> None: ...
             def HasField(
-                self, field_name: typing_extensions.Literal["value", b"value"]
+                self, field_name: typing.Literal["value", b"value"]
             ) -> builtins.bool: ...
             def ClearField(
-                self,
-                field_name: typing_extensions.Literal["key", b"key", "value", b"value"],
+                self, field_name: typing.Literal["key", b"key", "value", b"value"]
             ) -> None: ...
@@ -752,8 +751,6 @@
         CONTENTS_FIELD_NUMBER: builtins.int
-        name: typing.Text = ...
+        name: builtins.str
         """The tensor name."""
-
-        datatype: typing.Text = ...
+        datatype: builtins.str
         """The tensor data type."""
-
         @property
@@ -766,3 +763,2 @@ def shape(
             """The tensor shape."""
-            pass
 
@@ -773,9 +769,8 @@ @property
         ) -> google.protobuf.internal.containers.MessageMap[
-            typing.Text, global___InferParameter
+            builtins.str, Global___InferParameter
         ]:
             """Optional output tensor parameters."""
-            pass
 
         @property
-        def contents(self) -> global___InferTensorContents:
+        def contents(self) -> Global___InferTensorContents:
             """The output tensor data.
@@ -785,3 +780,2 @@ This field must not be specified if tensor contents are being specified
             """
-            pass
 
@@ -791,16 +785,16 @@ def __init__(
             *,
-            name: typing.Text = ...,
-            datatype: typing.Text = ...,
-            shape: typing.Optional[typing.Iterable[builtins.int]] = ...,
-            parameters: typing.Optional[
-                typing.Mapping[typing.Text, global___InferParameter]
-            ] = ...,
-            contents: typing.Optional[global___InferTensorContents] = ...,
+            name: builtins.str = ...,
+            datatype: builtins.str = ...,
+            shape: collections.abc.Iterable[builtins.int] | None = ...,
+            parameters: (
+                collections.abc.Mapping[builtins.str, Global___InferParameter] | None
+            ) = ...,
+            contents: Global___InferTensorContents | None = ...,
         ) -> None: ...
         def HasField(
-            self, field_name: typing_extensions.Literal["contents", b"contents"]
+            self, field_name: typing.Literal["contents", b"contents"]
         ) -> builtins.bool: ...
         def ClearField(
             self,
-            field_name: typing_extensions.Literal[
+            field_name: typing.Literal[
                 "contents",
@@ -819,21 +813,22 @@ b"contents",
 
+    @typing.final
     class ParametersEntry(google.protobuf.message.Message):
-        DESCRIPTOR: google.protobuf.descriptor.Descriptor = ...
+        DESCRIPTOR: google.protobuf.descriptor.Descriptor
+
         KEY_FIELD_NUMBER: builtins.int
         VALUE_FIELD_NUMBER: builtins.int
-        key: typing.Text = ...
+        key: builtins.str
         @property
-        def value(self) -> global___InferParameter: ...
+        def value(self) -> Global___InferParameter: ...
         def __init__(
             self,
             *,
-            key: typing.Text = ...,
-            value: typing.Optional[global___InferParameter] = ...,
+            key: builtins.str = ...,
+            value: Global___InferParameter | None = ...,
         ) -> None: ...
         def HasField(
-            self, field_name: typing_extensions.Literal["value", b"value"]
+            self, field_name: typing.Literal["value", b"value"]
         ) -> builtins.bool: ...
         def ClearField(
-            self,
-            field_name: typing_extensions.Literal["key", b"key", "value", b"value"],
+            self, field_name: typing.Literal["key", b"key", "value", b"value"]
         ) -> None: ...
@@ -847,11 +842,8 @@
     RAW_OUTPUT_CONTENTS_FIELD_NUMBER: builtins.int
-    model_name: typing.Text = ...
+    model_name: builtins.str
     """The name of the model used for inference."""
-
-    model_version: typing.Text = ...
+    model_version: builtins.str
     """The version of the model used for inference."""
-
-    id: typing.Text = ...
+    id: builtins.str
     """The id of the inference request if one was specified."""
-
     @property
@@ -861,6 +853,5 @@ def parameters(
     ) -> google.protobuf.internal.containers.MessageMap[
-        typing.Text, global___InferParameter
+        builtins.str, Global___InferParameter
     ]:
         """Optional inference response parameters."""
-        pass
 
@@ -871,6 +862,5 @@ @property
     ) -> google.protobuf.internal.containers.RepeatedCompositeFieldContainer[
-        global___ModelInferResponse.InferOutputTensor
+        Global___ModelInferResponse.InferOutputTensor
     ]:
         """The output tensors holding inference results."""
-        pass
 
@@ -901,3 +891,2 @@ @property
         """
-        pass
 
@@ -907,16 +896,17 @@ def __init__(
         *,
-        model_name: typing.Text = ...,
-        model_version: typing.Text = ...,
-        id: typing.Text = ...,
-        parameters: typing.Optional[
-            typing.Mapping[typing.Text, global___InferParameter]
-        ] = ...,
-        outputs: typing.Optional[
-            typing.Iterable[global___ModelInferResponse.InferOutputTensor]
-        ] = ...,
-        raw_output_contents: typing.Optional[typing.Iterable[builtins.bytes]] = ...,
+        model_name: builtins.str = ...,
+        model_version: builtins.str = ...,
+        id: builtins.str = ...,
+        parameters: (
+            collections.abc.Mapping[builtins.str, Global___InferParameter] | None
+        ) = ...,
+        outputs: (
+            collections.abc.Iterable[Global___ModelInferResponse.InferOutputTensor]
+            | None
+        ) = ...,
+        raw_output_contents: collections.abc.Iterable[builtins.bytes] | None = ...,
     ) -> None: ...
     def ClearField(
         self,
-        field_name: typing_extensions.Literal[
+        field_name: typing.Literal[
             "id",
@@ -937,4 +927,5 @@ b"id",
 
-global___ModelInferResponse = ModelInferResponse
+Global___ModelInferResponse: typing_extensions.TypeAlias = ModelInferResponse
 
[e-mail]
 class InferParameter(google.protobuf.message.Message):
@@ -945,15 +936,13 @@ """
 
-    DESCRIPTOR: google.protobuf.descriptor.Descriptor = ...
+    DESCRIPTOR: google.protobuf.descriptor.Descriptor
+
     BOOL_PARAM_FIELD_NUMBER: builtins.int
     INT64_PARAM_FIELD_NUMBER: builtins.int
     STRING_PARAM_FIELD_NUMBER: builtins.int
-    bool_param: builtins.bool = ...
+    bool_param: builtins.bool
     """A boolean parameter value."""
-
-    int64_param: builtins.int = ...
+    int64_param: builtins.int
     """An int64 parameter value."""
-
-    string_param: typing.Text = ...
+    string_param: builtins.str
     """A string parameter value."""
-
     def __init__(
@@ -964,7 +953,7 @@ self,
         int64_param: builtins.int = ...,
-        string_param: typing.Text = ...,
+        string_param: builtins.str = ...,
     ) -> None: ...
     def HasField(
         self,
-        field_name: typing_extensions.Literal[
+        field_name: typing.Literal[
             "bool_param",
@@ -982,3 +971,3 @@ b"bool_param",
         self,
-        field_name: typing_extensions.Literal[
+        field_name: typing.Literal[
             "bool_param",
@@ -995,10 +984,8 @@ b"bool_param",
     def WhichOneof(
-        self,
-        oneof_group: typing_extensions.Literal["parameter_choice", b"parameter_choice"],
-    ) -> typing.Optional[
-        typing_extensions.Literal["bool_param", "int64_param", "string_param"]
-    ]: ...
+        self, oneof_group: typing.Literal["parameter_choice", b"parameter_choice"]
+    ) -> typing.Literal["bool_param", "int64_param", "string_param"] | None: ...
 
-global___InferParameter = InferParameter
+Global___InferParameter: typing_extensions.TypeAlias = InferParameter
 
[e-mail]
 class InferTensorContents(google.protobuf.message.Message):
@@ -1012,3 +999,4 @@ """
 
-    DESCRIPTOR: google.protobuf.descriptor.Descriptor = ...
+    DESCRIPTOR: google.protobuf.descriptor.Descriptor
+
     BOOL_CONTENTS_FIELD_NUMBER: builtins.int
@@ -1032,3 +1020,2 @@ INT_CONTENTS_FIELD_NUMBER: builtins.int
         """
-        pass
 
@@ -1044,3 +1031,2 @@ @property
         """
-        pass
 
@@ -1055,3 +1041,2 @@ @property
         """
-        pass
 
@@ -1067,3 +1052,2 @@ @property
         """
-        pass
 
@@ -1078,3 +1062,2 @@ @property
         """
-        pass
 
@@ -1091,3 +1074,2 @@ @property
         """
-        pass
 
@@ -1104,3 +1086,2 @@ @property
         """
-        pass
 
@@ -1117,3 +1098,2 @@ @property
         """
-        pass
 
@@ -1123,14 +1103,14 @@ def __init__(
         *,
-        bool_contents: typing.Optional[typing.Iterable[builtins.bool]] = ...,
-        int_contents: typing.Optional[typing.Iterable[builtins.int]] = ...,
-        int64_contents: typing.Optional[typing.Iterable[builtins.int]] = ...,
-        uint_contents: typing.Optional[typing.Iterable[builtins.int]] = ...,
-        uint64_contents: typing.Optional[typing.Iterable[builtins.int]] = ...,
-        fp32_contents: typing.Optional[typing.Iterable[builtins.float]] = ...,
-        fp64_contents: typing.Optional[typing.Iterable[builtins.float]] = ...,
-        bytes_contents: typing.Optional[typing.Iterable[builtins.bytes]] = ...,
+        bool_contents: collections.abc.Iterable[builtins.bool] | None = ...,
+        int_contents: collections.abc.Iterable[builtins.int] | None = ...,
+        int64_contents: collections.abc.Iterable[builtins.int] | None = ...,
+        uint_contents: collections.abc.Iterable[builtins.int] | None = ...,
+        uint64_contents: collections.abc.Iterable[builtins.int] | None = ...,
+        fp32_contents: collections.abc.Iterable[builtins.float] | None = ...,
+        fp64_contents: collections.abc.Iterable[builtins.float] | None = ...,
+        bytes_contents: collections.abc.Iterable[builtins.bytes] | None = ...,
     ) -> None: ...
     def ClearField(
         self,
-        field_name: typing_extensions.Literal[
+        field_name: typing.Literal[
             "bool_contents",
@@ -1155,4 +1135,5 @@ b"bool_contents",
 
-global___InferTensorContents = InferTensorContents
+Global___InferTensorContents: typing_extensions.TypeAlias = InferTensorContents
 
[e-mail]
 class ModelRepositoryParameter(google.protobuf.message.Message):
@@ -1169,3 +1150,4 @@ """
 
-    DESCRIPTOR: google.protobuf.descriptor.Descriptor = ...
+    DESCRIPTOR: google.protobuf.descriptor.Descriptor
+
     BOOL_PARAM_FIELD_NUMBER: builtins.int
@@ -1175,14 +1157,10 @@ INT64_PARAM_FIELD_NUMBER: builtins.int
     BYTES_PARAM_FIELD_NUMBER: builtins.int
-    bool_param: builtins.bool = ...
+    bool_param: builtins.bool
     """A boolean parameter value."""
-
-    int64_param: builtins.int = ...
+    int64_param: builtins.int
     """An int64 parameter value."""
-
-    string_param: typing.Text = ...
+    string_param: builtins.str
     """A string parameter value."""
-
-    bytes_param: builtins.bytes = ...
+    bytes_param: builtins.bytes
     """A bytes parameter value."""
-
     def __init__(
@@ -1193,3 +1171,3 @@ self,
         int64_param: builtins.int = ...,
-        string_param: typing.Text = ...,
+        string_param: builtins.str = ...,
         bytes_param: builtins.bytes = ...,
@@ -1199,3 +1177,3 @@ ) -> None: ...
         self,
-        field_name: typing_extensions.Literal[
+        field_name: typing.Literal[
             "bool_param",
@@ -1215,3 +1193,3 @@ b"bool_param",
         self,
-        field_name: typing_extensions.Literal[
+        field_name: typing.Literal[
             "bool_param",
@@ -1230,28 +1208,28 @@ b"bool_param",
     def WhichOneof(
-        self,
-        oneof_group: typing_extensions.Literal["parameter_choice", b"parameter_choice"],
-    ) -> typing.Optional[
-        typing_extensions.Literal[
-            "bool_param", "int64_param", "string_param", "bytes_param"
-        ]
-    ]: ...
+        self, oneof_group: typing.Literal["parameter_choice", b"parameter_choice"]
+    ) -> (
+        typing.Literal["bool_param", "int64_param", "string_param", "bytes_param"]
+        | None
+    ): ...
 
-global___ModelRepositoryParameter = ModelRepositoryParameter
+Global___ModelRepositoryParameter: typing_extensions.TypeAlias = (
+    ModelRepositoryParameter
+)
 
[e-mail]
 class RepositoryIndexRequest(google.protobuf.message.Message):
-    DESCRIPTOR: google.protobuf.descriptor.Descriptor = ...
+    DESCRIPTOR: google.protobuf.descriptor.Descriptor
+
     REPOSITORY_NAME_FIELD_NUMBER: builtins.int
     READY_FIELD_NUMBER: builtins.int
-    repository_name: typing.Text = ...
+    repository_name: builtins.str
     """The name of the repository. If empty the index is returned
     for all repositories.
     """
-
-    ready: builtins.bool = ...
+    ready: builtins.bool
     """If true return only models currently ready for inferencing."""
-
     def __init__(
         self,
         *,
-        repository_name: typing.Text = ...,
+        repository_name: builtins.str = ...,
         ready: builtins.bool = ...,
@@ -1261,3 +1239,3 @@ ) -> None: ...
         self,
-        field_name: typing_extensions.Literal[
+        field_name: typing.Literal[
             "ready", b"ready", "repository_name", b"repository_name"
@@ -1267,11 +1245,14 @@ ],
 
-global___RepositoryIndexRequest = RepositoryIndexRequest
+Global___RepositoryIndexRequest: typing_extensions.TypeAlias = RepositoryIndexRequest
 
[e-mail]
 class RepositoryIndexResponse(google.protobuf.message.Message):
-    DESCRIPTOR: google.protobuf.descriptor.Descriptor = ...
+    DESCRIPTOR: google.protobuf.descriptor.Descriptor
 
+    @typing.final
     class ModelIndex(google.protobuf.message.Message):
         """Index entry for a model."""
 
-        DESCRIPTOR: google.protobuf.descriptor.Descriptor = ...
+        DESCRIPTOR: google.protobuf.descriptor.Descriptor
+
         NAME_FIELD_NUMBER: builtins.int
@@ -1281,25 +1262,21 @@ VERSION_FIELD_NUMBER: builtins.int
         REASON_FIELD_NUMBER: builtins.int
-        name: typing.Text = ...
+        name: builtins.str
         """The name of the model."""
-
-        version: typing.Text = ...
+        version: builtins.str
         """The version of the model."""
-
-        state: typing.Text = ...
+        state: builtins.str
         """The state of the model."""
-
-        reason: typing.Text = ...
+        reason: builtins.str
         """The reason, if any, that the model is in the given state."""
-
         def __init__(
             self,
             *,
-            name: typing.Text = ...,
-            version: typing.Text = ...,
-            state: typing.Text = ...,
-            reason: typing.Text = ...,
+            name: builtins.str = ...,
+            version: builtins.str = ...,
+            state: builtins.str = ...,
+            reason: builtins.str = ...,
         ) -> None: ...
         def ClearField(
             self,
-            field_name: typing_extensions.Literal[
+            field_name: typing.Literal[
                 "name",
@@ -1321,6 +1298,5 @@ b"name",
     ) -> google.protobuf.internal.containers.RepeatedCompositeFieldContainer[
-        global___RepositoryIndexResponse.ModelIndex
+        Global___RepositoryIndexResponse.ModelIndex
     ]:
         """An index entry for each model."""
-        pass
 
@@ -1330,34 +1306,34 @@ def __init__(
         *,
-        models: typing.Optional[
-            typing.Iterable[global___RepositoryIndexResponse.ModelIndex]
-        ] = ...,
+        models: (
+            collections.abc.Iterable[Global___RepositoryIndexResponse.ModelIndex] | None
+        ) = ...,
     ) -> None: ...
-    def ClearField(
-        self, field_name: typing_extensions.Literal["models", b"models"]
-    ) -> None: ...
+    def ClearField(self, field_name: typing.Literal["models", b"models"]) -> None: ...
 
-global___RepositoryIndexResponse = RepositoryIndexResponse
+Global___RepositoryIndexResponse: typing_extensions.TypeAlias = RepositoryIndexResponse
 
[e-mail]
 class RepositoryModelLoadRequest(google.protobuf.message.Message):
-    DESCRIPTOR: google.protobuf.descriptor.Descriptor = ...
+    DESCRIPTOR: google.protobuf.descriptor.Descriptor
 
+    @typing.final
     class ParametersEntry(google.protobuf.message.Message):
-        DESCRIPTOR: google.protobuf.descriptor.Descriptor = ...
+        DESCRIPTOR: google.protobuf.descriptor.Descriptor
+
         KEY_FIELD_NUMBER: builtins.int
         VALUE_FIELD_NUMBER: builtins.int
-        key: typing.Text = ...
+        key: builtins.str
         @property
-        def value(self) -> global___ModelRepositoryParameter: ...
+        def value(self) -> Global___ModelRepositoryParameter: ...
         def __init__(
             self,
             *,
-            key: typing.Text = ...,
-            value: typing.Optional[global___ModelRepositoryParameter] = ...,
+            key: builtins.str = ...,
+            value: Global___ModelRepositoryParameter | None = ...,
         ) -> None: ...
         def HasField(
-            self, field_name: typing_extensions.Literal["value", b"value"]
+            self, field_name: typing.Literal["value", b"value"]
         ) -> builtins.bool: ...
         def ClearField(
-            self,
-            field_name: typing_extensions.Literal["key", b"key", "value", b"value"],
+            self, field_name: typing.Literal["key", b"key", "value", b"value"]
         ) -> None: ...
@@ -1368,10 +1344,8 @@
     PARAMETERS_FIELD_NUMBER: builtins.int
-    repository_name: typing.Text = ...
+    repository_name: builtins.str
     """The name of the repository to load from. If empty the model
     is loaded from any repository.
     """
-
-    model_name: typing.Text = ...
+    model_name: builtins.str
     """The name of the model to load, or reload."""
-
     @property
@@ -1381,6 +1355,5 @@ def parameters(
     ) -> google.protobuf.internal.containers.MessageMap[
-        typing.Text, global___ModelRepositoryParameter
+        builtins.str, Global___ModelRepositoryParameter
     ]:
         """Optional model repository request parameters."""
-        pass
 
@@ -1390,11 +1363,12 @@ def __init__(
         *,
-        repository_name: typing.Text = ...,
-        model_name: typing.Text = ...,
-        parameters: typing.Optional[
-            typing.Mapping[typing.Text, global___ModelRepositoryParameter]
-        ] = ...,
+        repository_name: builtins.str = ...,
+        model_name: builtins.str = ...,
+        parameters: (
+            collections.abc.Mapping[builtins.str, Global___ModelRepositoryParameter]
+            | None
+        ) = ...,
     ) -> None: ...
     def ClearField(
         self,
-        field_name: typing_extensions.Literal[
+        field_name: typing.Literal[
             "model_name",
@@ -1409,6 +1383,10 @@ b"model_name",
 
-global___RepositoryModelLoadRequest = RepositoryModelLoadRequest
+Global___RepositoryModelLoadRequest: typing_extensions.TypeAlias = (
+    RepositoryModelLoadRequest
+)
 
[e-mail]
 class RepositoryModelLoadResponse(google.protobuf.message.Message):
-    DESCRIPTOR: google.protobuf.descriptor.Descriptor = ...
+    DESCRIPTOR: google.protobuf.descriptor.Descriptor
+
     def __init__(
@@ -1418,26 +1396,30 @@ self,
 
-global___RepositoryModelLoadResponse = RepositoryModelLoadResponse
+Global___RepositoryModelLoadResponse: typing_extensions.TypeAlias = (
+    RepositoryModelLoadResponse
+)
 
[e-mail]
 class RepositoryModelUnloadRequest(google.protobuf.message.Message):
-    DESCRIPTOR: google.protobuf.descriptor.Descriptor = ...
+    DESCRIPTOR: google.protobuf.descriptor.Descriptor
 
+    @typing.final
     class ParametersEntry(google.protobuf.message.Message):
-        DESCRIPTOR: google.protobuf.descriptor.Descriptor = ...
+        DESCRIPTOR: google.protobuf.descriptor.Descriptor
+
         KEY_FIELD_NUMBER: builtins.int
         VALUE_FIELD_NUMBER: builtins.int
-        key: typing.Text = ...
+        key: builtins.str
         @property
-        def value(self) -> global___ModelRepositoryParameter: ...
+        def value(self) -> Global___ModelRepositoryParameter: ...
         def __init__(
             self,
             *,
-            key: typing.Text = ...,
-            value: typing.Optional[global___ModelRepositoryParameter] = ...,
+            key: builtins.str = ...,
+            value: Global___ModelRepositoryParameter | None = ...,
         ) -> None: ...
         def HasField(
-            self, field_name: typing_extensions.Literal["value", b"value"]
+            self, field_name: typing.Literal["value", b"value"]
         ) -> builtins.bool: ...
         def ClearField(
-            self,
-            field_name: typing_extensions.Literal["key", b"key", "value", b"value"],
+            self, field_name: typing.Literal["key", b"key", "value", b"value"]
         ) -> None: ...
@@ -1448,10 +1430,8 @@
     PARAMETERS_FIELD_NUMBER: builtins.int
-    repository_name: typing.Text = ...
+    repository_name: builtins.str
     """The name of the repository from which the model was originally
     loaded. If empty the repository is not considered.
     """
-
-    model_name: typing.Text = ...
+    model_name: builtins.str
     """The name of the model to unload."""
-
     @property
@@ -1461,6 +1441,5 @@ def parameters(
     ) -> google.protobuf.internal.containers.MessageMap[
-        typing.Text, global___ModelRepositoryParameter
+        builtins.str, Global___ModelRepositoryParameter
     ]:
         """Optional model repository request parameters."""
-        pass
 
@@ -1470,11 +1449,12 @@ def __init__(
         *,
-        repository_name: typing.Text = ...,
-        model_name: typing.Text = ...,
-        parameters: typing.Optional[
-            typing.Mapping[typing.Text, global___ModelRepositoryParameter]
-        ] = ...,
+        repository_name: builtins.str = ...,
+        model_name: builtins.str = ...,
+        parameters: (
+            collections.abc.Mapping[builtins.str, Global___ModelRepositoryParameter]
+            | None
+        ) = ...,
     ) -> None: ...
     def ClearField(
         self,
-        field_name: typing_extensions.Literal[
+        field_name: typing.Literal[
             "model_name",
@@ -1489,6 +1469,10 @@ b"model_name",
 
-global___RepositoryModelUnloadRequest = RepositoryModelUnloadRequest
+Global___RepositoryModelUnloadRequest: typing_extensions.TypeAlias = (
+    RepositoryModelUnloadRequest
+)
 
[e-mail]
 class RepositoryModelUnloadResponse(google.protobuf.message.Message):
-    DESCRIPTOR: google.protobuf.descriptor.Descriptor = ...
+    DESCRIPTOR: google.protobuf.descriptor.Descriptor
+
     def __init__(
@@ -1498,2 +1482,4 @@ self,
 
-global___RepositoryModelUnloadResponse = RepositoryModelUnloadResponse
+Global___RepositoryModelUnloadResponse: typing_extensions.TypeAlias = (
+    RepositoryModelUnloadResponse
+)

mlserver/grpc/model_repository_pb2.pyi

@@ -7,26 +7,32 @@ """
 import builtins
+import collections.abc
 import google.protobuf.descriptor
 import google.protobuf.internal.containers
 import google.protobuf.message
+import sys
 import typing
-import typing_extensions
 
-DESCRIPTOR: google.protobuf.descriptor.FileDescriptor = ...
+if sys.version_info >= (3, 10):
+    import typing as typing_extensions
+else:
+    import typing_extensions
 
+DESCRIPTOR: google.protobuf.descriptor.FileDescriptor
+
[e-mail]
 class RepositoryIndexRequest(google.protobuf.message.Message):
-    DESCRIPTOR: google.protobuf.descriptor.Descriptor = ...
+    DESCRIPTOR: google.protobuf.descriptor.Descriptor
+
     REPOSITORY_NAME_FIELD_NUMBER: builtins.int
     READY_FIELD_NUMBER: builtins.int
-    repository_name: typing.Text = ...
+    repository_name: builtins.str
     """The name of the repository. If empty the index is returned
     for all repositories.
     """
-
-    ready: builtins.bool = ...
+    ready: builtins.bool
     """If true return only models currently ready for inferencing."""
-
     def __init__(
         self,
         *,
-        repository_name: typing.Text = ...,
+        repository_name: builtins.str = ...,
         ready: builtins.bool = ...,
@@ -36,3 +42,3 @@ ) -> None: ...
         self,
-        field_name: typing_extensions.Literal[
+        field_name: typing.Literal[
             "ready", b"ready", "repository_name", b"repository_name"
@@ -42,11 +48,14 @@ ],
 
-global___RepositoryIndexRequest = RepositoryIndexRequest
+Global___RepositoryIndexRequest: typing_extensions.TypeAlias = RepositoryIndexRequest
 
[e-mail]
 class RepositoryIndexResponse(google.protobuf.message.Message):
-    DESCRIPTOR: google.protobuf.descriptor.Descriptor = ...
+    DESCRIPTOR: google.protobuf.descriptor.Descriptor
 
+    @typing.final
     class ModelIndex(google.protobuf.message.Message):
         """Index entry for a model."""
 
-        DESCRIPTOR: google.protobuf.descriptor.Descriptor = ...
+        DESCRIPTOR: google.protobuf.descriptor.Descriptor
+
         NAME_FIELD_NUMBER: builtins.int
@@ -56,25 +65,21 @@ VERSION_FIELD_NUMBER: builtins.int
         REASON_FIELD_NUMBER: builtins.int
-        name: typing.Text = ...
+        name: builtins.str
         """The name of the model."""
-
-        version: typing.Text = ...
+        version: builtins.str
         """The version of the model."""
-
-        state: typing.Text = ...
+        state: builtins.str
         """The state of the model."""
-
-        reason: typing.Text = ...
+        reason: builtins.str
         """The reason, if any, that the model is in the given state."""
-
         def __init__(
             self,
             *,
-            name: typing.Text = ...,
-            version: typing.Text = ...,
-            state: typing.Text = ...,
-            reason: typing.Text = ...,
+            name: builtins.str = ...,
+            version: builtins.str = ...,
+            state: builtins.str = ...,
+            reason: builtins.str = ...,
         ) -> None: ...
         def ClearField(
             self,
-            field_name: typing_extensions.Literal[
+            field_name: typing.Literal[
                 "name",
@@ -96,6 +101,5 @@ b"name",
     ) -> google.protobuf.internal.containers.RepeatedCompositeFieldContainer[
-        global___RepositoryIndexResponse.ModelIndex
+        Global___RepositoryIndexResponse.ModelIndex
     ]:
         """An index entry for each model."""
-        pass
 
@@ -105,33 +109,31 @@ def __init__(
         *,
-        models: typing.Optional[
-            typing.Iterable[global___RepositoryIndexResponse.ModelIndex]
-        ] = ...,
+        models: (
+            collections.abc.Iterable[Global___RepositoryIndexResponse.ModelIndex] | None
+        ) = ...,
     ) -> None: ...
-    def ClearField(
-        self, field_name: typing_extensions.Literal["models", b"models"]
-    ) -> None: ...
+    def ClearField(self, field_name: typing.Literal["models", b"models"]) -> None: ...
 
-global___RepositoryIndexResponse = RepositoryIndexResponse
+Global___RepositoryIndexResponse: typing_extensions.TypeAlias = RepositoryIndexResponse
 
[e-mail]
 class RepositoryModelLoadRequest(google.protobuf.message.Message):
-    DESCRIPTOR: google.protobuf.descriptor.Descriptor = ...
+    DESCRIPTOR: google.protobuf.descriptor.Descriptor
+
     REPOSITORY_NAME_FIELD_NUMBER: builtins.int
     MODEL_NAME_FIELD_NUMBER: builtins.int
-    repository_name: typing.Text = ...
+    repository_name: builtins.str
     """The name of the repository to load from. If empty the model
     is loaded from any repository.
     """
-
-    model_name: typing.Text = ...
+    model_name: builtins.str
     """The name of the model to load, or reload."""
-
     def __init__(
         self,
         *,
-        repository_name: typing.Text = ...,
-        model_name: typing.Text = ...,
+        repository_name: builtins.str = ...,
+        model_name: builtins.str = ...,
     ) -> None: ...
     def ClearField(
         self,
-        field_name: typing_extensions.Literal[
+        field_name: typing.Literal[
             "model_name", b"model_name", "repository_name", b"repository_name"
@@ -141,6 +143,10 @@ ],
 
-global___RepositoryModelLoadRequest = RepositoryModelLoadRequest
+Global___RepositoryModelLoadRequest: typing_extensions.TypeAlias = (
+    RepositoryModelLoadRequest
+)
 
[e-mail]
 class RepositoryModelLoadResponse(google.protobuf.message.Message):
-    DESCRIPTOR: google.protobuf.descriptor.Descriptor = ...
+    DESCRIPTOR: google.protobuf.descriptor.Descriptor
+
     def __init__(
@@ -150,25 +156,27 @@ self,
 
-global___RepositoryModelLoadResponse = RepositoryModelLoadResponse
+Global___RepositoryModelLoadResponse: typing_extensions.TypeAlias = (
+    RepositoryModelLoadResponse
+)
 
[e-mail]
 class RepositoryModelUnloadRequest(google.protobuf.message.Message):
-    DESCRIPTOR: google.protobuf.descriptor.Descriptor = ...
+    DESCRIPTOR: google.protobuf.descriptor.Descriptor
+
     REPOSITORY_NAME_FIELD_NUMBER: builtins.int
     MODEL_NAME_FIELD_NUMBER: builtins.int
-    repository_name: typing.Text = ...
+    repository_name: builtins.str
     """The name of the repository from which the model was originally
     loaded. If empty the repository is not considered.
     """
-
-    model_name: typing.Text = ...
+    model_name: builtins.str
     """The name of the model to unload."""
-
     def __init__(
         self,
         *,
-        repository_name: typing.Text = ...,
-        model_name: typing.Text = ...,
+        repository_name: builtins.str = ...,
+        model_name: builtins.str = ...,
     ) -> None: ...
     def ClearField(
         self,
-        field_name: typing_extensions.Literal[
+        field_name: typing.Literal[
             "model_name", b"model_name", "repository_name", b"repository_name"
@@ -178,6 +186,10 @@ ],
 
-global___RepositoryModelUnloadRequest = RepositoryModelUnloadRequest
+Global___RepositoryModelUnloadRequest: typing_extensions.TypeAlias = (
+    RepositoryModelUnloadRequest
+)
 
[e-mail]
 class RepositoryModelUnloadResponse(google.protobuf.message.Message):
-    DESCRIPTOR: google.protobuf.descriptor.Descriptor = ...
+    DESCRIPTOR: google.protobuf.descriptor.Descriptor
+
     def __init__(
@@ -187,2 +199,4 @@ self,
 
-global___RepositoryModelUnloadResponse = RepositoryModelUnloadResponse
+Global___RepositoryModelUnloadResponse: typing_extensions.TypeAlias = (
+    RepositoryModelUnloadResponse
+)

mlserver/metrics/prometheus.py

@@ -34,2 +34,9 @@ import glob
 
+async def try_remove(filename: str):
+    try:
+        await remove(filename)
+    except FileNotFoundError:
+        pass
+
+
 async def stop_metrics(settings: Settings, pid: int):
@@ -42,3 +49,4 @@ if not settings.parallel_workers:
     matching_files = glob.glob(pattern)
-    await asyncio.gather(*[remove(f) for f in matching_files])
+    if len(matching_files) > 0:
+        await asyncio.gather(*[try_remove(f) for f in matching_files])
 
@@ -45,0 +53,0 @@

mlserver/parallel/dispatcher.py

@@ -198,3 +198,3 @@ import asyncio
         logger.debug("Starting response processing loop...")
-        loop = asyncio.get_event_loop()
+        loop = asyncio.get_running_loop()
         while self._active:
@@ -201,0 +201,0 @@ response = await loop.run_in_executor(self._executor, self._responses.get)

mlserver/parallel/registry.py

@@ -202,3 +202,3 @@ import asyncio
                 "settings. "
-                "This field will be removed in MLServer 1.2.0. "
+                "This field will be removed in MLServer 1.8.0. "
                 "To access the new field, you can either update the "
@@ -205,0 +205,0 @@ "`settings.json` file, or update the `MLSERVER_PARALLEL_WORKERS` "

mlserver/parallel/worker.py

@@ -12,3 +12,3 @@ import asyncio
 from ..registry import MultiModelRegistry
-from ..utils import install_uvloop_event_loop, schedule_with_callback
+from ..utils import AsyncManager, schedule_with_callback
 from ..logging import configure_logger
@@ -68,9 +68,8 @@ from ..settings import Settings
         with ctx:
-            install_uvloop_event_loop()
+            async_mgr = AsyncManager(self._ignore_signals)
             configure_logger(self._settings)
             configure_metrics(self._settings)
-            self._ignore_signals()
-            asyncio.run(self.coro_run())
+            async_mgr.run(self.coro_run())
 
-    def _ignore_signals(self):
+    def _ignore_signals(self, loop):
         """
@@ -83,4 +82,2 @@ Uvloop will try to propagate the main process' signals to the
         """
-        loop = asyncio.get_event_loop()
-
         for sign in IGNORED_SIGNALS:
@@ -100,3 +97,3 @@ # Ensure that signal handlers are a no-op, to let the main process
         self.__inner_init__()
-        loop = asyncio.get_event_loop()
+        loop = asyncio.get_running_loop()
         while self._active:
@@ -103,0 +100,0 @@ readable = await loop.run_in_executor(self._executor, self._select)

mlserver/rest/app.py

@@ -32,3 +32,3 @@ from typing import Callable
         response_class=Response,
-        **kwargs
+        **kwargs,
     ):
@@ -40,3 +40,3 @@ super().__init__(
             response_class=Response,
-            **kwargs
+            **kwargs,
         )
@@ -43,0 +43,0 @@

mlserver/server.py

@@ -169,3 +169,3 @@ import asyncio
     def _add_signal_handlers(self):
-        loop = asyncio.get_event_loop()
+        loop = asyncio.get_running_loop()
 
@@ -172,0 +172,0 @@ for sig in HANDLED_SIGNALS:

mlserver/settings.py

@@ -168,3 +168,3 @@ import sys
 
-    debug: bool = True
+    debug: bool = False
 
@@ -444,3 +444,3 @@ parallel_workers: int = DEFAULT_PARALLEL_WORKERS
     parallel_workers: Optional[int] = Field(
-        None,
+        default=None,
         deprecated=True,
@@ -447,0 +447,0 @@ description=(

mlserver/utils.py

@@ -0,8 +1,19 @@
+# mypy: disable-error-code="call-arg,attr-defined"
+#
+# The typechecking exceptions referenced above are needed because code in this file
+# dynamically considers different python versions, and some called functions are
+# deprecated or take different arguments across versions.
+
 import os
+import sys
 import uuid
 import asyncio
+from enum import Enum
+from typing import cast
+import warnings
 import urllib.parse
 
 from asyncio import Task
-from typing import Callable, Dict, Optional, List
+from typing import Any, Optional, TypeVar
+from collections.abc import Callable, Coroutine
 
@@ -14,5 +25,173 @@ from .logging import logger
 
+_CoroRetT = TypeVar("_CoroRetT")
+EventLoopSignalHandleConfigFn = Callable[[asyncio.AbstractEventLoop], None]
+_RUN_IN_LOOP_ERROR = (
+    "AsyncManager::run() cannot be called from within an existing event loop"
+)
 
+
+class EventLoopBackend(Enum):
+    ASYNCIO_DEFAULT = 1
+    UVLOOP = 2
+
+
+class AsyncManager:
+    def __init__(self, signal_handle_config_fn=None):
+        self._event_loop_backend = EventLoopBackend.ASYNCIO_DEFAULT
+        self._signal_handle_config_fn = signal_handle_config_fn
+        try:
+            import uvloop
+
+            self._event_loop_backend = EventLoopBackend.UVLOOP
+            self._async_run_fn = uvloop.run
+        except ImportError:
+            # else keep the standard asyncio loop as a fallback
+            self._async_run_fn = asyncio.run
+            logger.warning(
+                "uvloop not available, falling back on default asyncio runner"
+            )
+
+    def _loop_factory(self) -> asyncio.AbstractEventLoop:
+        new_loop: asyncio.AbstractEventLoop
+        if self._event_loop_backend == EventLoopBackend.UVLOOP:
+            import uvloop
+
+            new_loop = uvloop.new_event_loop()
+        else:
+            new_loop = asyncio.new_event_loop()
+
+        if self._signal_handle_config_fn is not None:
+            self._signal_handle_config_fn(new_loop)
+
+        return new_loop
+
+    @property
+    def event_loop_backend(self) -> EventLoopBackend:
+        return self._event_loop_backend
+
+    @property
+    def event_loop_policy(self) -> asyncio.AbstractEventLoopPolicy:
+        """
+        In python versions < 3.16, return the event loop policy.
+        It prefers the policy returned by uvloop where available.
+        Deprecated from python 3.12, but used within pytest tests until pytest-async
+        itself provides an alternative way to manage event loops across different
+        python versions.
+        """
+        python_version_info = sys.version_info[:2]
+        if python_version_info >= (3, 12):
+            warnings.warn(
+                "AsyncManager::event_loop_policy is deprecated since Python 3.12.",
+                DeprecationWarning,
+                stacklevel=1,
+            )
+        if python_version_info >= (3, 16):
+            raise RuntimeError("EventLoopPolicy no loger supported in Python >=3.16")
+        if self._event_loop_backend == EventLoopBackend.UVLOOP:
+            import uvloop
+
+            return uvloop.EventLoopPolicy()
+        else:
+            return cast(asyncio.AbstractEventLoopPolicy, asyncio.DefaultEventLoopPolicy)
+
+    def run(
+        self, coro: Coroutine[Any, Any, _CoroRetT], *, debug: Optional[bool] = None
+    ) -> _CoroRetT:
+        """
+        Run a coroutine in a new event loop, and close the loop afterwards.
+        Equivalent to asyncio.run(...) but with custom behaviour:
+
+            - The event loop created during the run call is backed by uvloop when
+            available, and falls back on standard the default asyncio implementation
+            otherwise.
+            - At the time of initialising the AsyncManager object, the resulting
+            loop can also be configured with custom handlers for os signals
+            (i.e SIGINT).
+
+        This function is compatible with different python versions from 3.9 to 3.16,
+        and papers over the differences in deprecated functions across these
+        versions.
+        """
+        if self._event_loop_backend == EventLoopBackend.UVLOOP:
+            # uvloop handles differences between python versions internally
+            return self._async_run_fn(coro, loop_factory=self._loop_factory)
+
+        python_version_info = sys.version_info[:2]
+
+        if python_version_info < (3, 11):
+            try:
+                asyncio.get_running_loop()
+                raise RuntimeError(_RUN_IN_LOOP_ERROR)
+            except RuntimeError:
+                pass
+
+            if not asyncio.iscoroutine(coro):
+                raise ValueError(f"expected coroutine, got {coro!r}")
+
+            loop = self._loop_factory()
+            try:
+                asyncio.set_event_loop(loop)
+                if debug is not None:
+                    loop.set_debug(debug)
+                return loop.run_until_complete(coro)
+            finally:
+                try:
+                    _cancel_all_tasks(loop)
+                    loop.run_until_complete(loop.shutdown_asyncgens())
+                    if hasattr(loop, "shutdown_default_executor"):
+                        loop.run_until_complete(loop.shutdown_default_executor())
+                finally:
+                    asyncio.set_event_loop(None)
+                    loop.close()
+
+        elif python_version_info == (3, 11):
+            try:
+                asyncio.get_running_loop()
+                raise RuntimeError(_RUN_IN_LOOP_ERROR)
+            except RuntimeError:
+                pass
+
+            with asyncio.Runner(
+                loop_factory=self._loop_factory,
+                debug=debug,
+            ) as runner:
+                return runner.run(coro)
+
+        else:
+            assert python_version_info > (3, 11)
+            # From python 3.12 onwards, asyncio.run() accepts a loop_factory argument
+            return asyncio.run(
+                coro,
+                debug=debug,
+                loop_factory=self._loop_factory,
+            )
+
+
+def _cancel_all_tasks(loop: asyncio.AbstractEventLoop) -> None:
+    # Copied from python/cpython
+
+    to_cancel = asyncio.all_tasks(loop)
+    if not to_cancel:
+        return
+
+    for task in to_cancel:
+        task.cancel()
+
+    loop.run_until_complete(asyncio.gather(*to_cancel, return_exceptions=True))
+
+    for task in to_cancel:
+        if task.cancelled():
+            continue
+        if task.exception() is not None:
+            loop.call_exception_handler(
+                {
+                    "message": "unhandled exception during asyncio.run() shutdown",
+                    "exception": task.exception(),
+                    "task": task,
+                }
+            )
+
+
 async def get_model_uri(
-    settings: ModelSettings, wellknown_filenames: List[str] = []
+    settings: ModelSettings, wellknown_filenames: list[str] = []
 ) -> str:
@@ -72,3 +251,3 @@ if not settings.parameters:
 def insert_headers(
-    inference_request: InferenceRequest, headers: Dict[str, str]
+    inference_request: InferenceRequest, headers: dict[str, str]
 ) -> InferenceRequest:
@@ -96,3 +275,3 @@ # Ensure parameters are present
 
-def extract_headers(inference_response: InferenceResponse) -> Optional[Dict[str, str]]:
+def extract_headers(inference_response: InferenceResponse) -> Optional[dict[str, str]]:
     if inference_response.parameters is None:
@@ -110,28 +289,2 @@ return None
 
-def _check_current_event_loop_policy() -> str:
-    policy = (
-        "uvloop"
-        if type(asyncio.get_event_loop_policy()).__module__.startswith("uvloop")
-        else "asyncio"
-    )
-    return policy
-
-
-def install_uvloop_event_loop():
-    if "uvloop" == _check_current_event_loop_policy():
-        return
-
-    try:
-        import uvloop
-
-        uvloop.install()
-    except ImportError:
-        # else keep the standard asyncio loop as a fallback
-        pass
-
-    policy = _check_current_event_loop_policy()
-
-    logger.debug(f"Using asyncio event-loop policy: {policy}")
-
-
 def schedule_with_callback(coro, cb) -> Task:
@@ -138,0 +291,0 @@ task = asyncio.create_task(coro)

mlserver/version.py

@@ -1,1 +0,1 @@
-__version__ = "1.7.1"
+__version__ = "1.7.2-rc1"

PKG-INFO

@@ -1,21 +0,16 @@
-Metadata-Version: 2.3
+Metadata-Version: 2.4
 Name: mlserver
-Version: 1.7.1
+Version: 1.7.2rc1
 Summary: MLServer
-License: Apache-2.0
+License-Expression: Apache-2.0
+License-File: LICENSE
 Author: Seldon Technologies Ltd.
 Author-email: [e-mail]
 Requires-Python: >=3.9,<3.13
-Classifier: License :: OSI Approved :: Apache Software License
+Classifier: Operating System :: POSIX
 Classifier: Operating System :: MacOS
-Classifier: Operating System :: POSIX
-Classifier: Programming Language :: Python :: 3
-Classifier: Programming Language :: Python :: 3.9
-Classifier: Programming Language :: Python :: 3.10
-Classifier: Programming Language :: Python :: 3.11
-Classifier: Programming Language :: Python :: 3.12
 Requires-Dist: aiofiles
 Requires-Dist: aiokafka
 Requires-Dist: click
-Requires-Dist: fastapi (>=0.88.0,!=0.89.0,<0.116.0)
+Requires-Dist: fastapi (>=0.88.0,!=0.89.0,<0.125.0)
 Requires-Dist: gevent
@@ -30,5 +25,5 @@ Requires-Dist: geventhttpclient
 Requires-Dist: opentelemetry-sdk (>=1.22.0,<2.0.0)
-Requires-Dist: orjson
+Requires-Dist: orjson (>=3.10,<4)
 Requires-Dist: pandas
-Requires-Dist: protobuf (>=5.27.2,<7.0.0)
+Requires-Dist: protobuf (>=5.29.5,<7.0.0)
 Requires-Dist: py-grpc-prometheus
@@ -40,5 +35,5 @@ Requires-Dist: pydantic (>=2.7.1,<3.0.0)
 Requires-Dist: starlette-exporter
-Requires-Dist: tritonclient[http] (>=2.42)
-Requires-Dist: uvicorn
-Requires-Dist: uvloop ; sys_platform != "win32" and sys_platform != "cygwin" and platform_python_implementation != "PyPy"
+Requires-Dist: tritonclient[http] (>=2.63)
+Requires-Dist: uvicorn (>=0.38.0)
+Requires-Dist: uvloop (>=0.22.1,<0.24) ; sys_platform != "win32" and sys_platform != "cygwin" and platform_python_implementation != "PyPy"
 Description-Content-Type: text/markdown
@@ -45,0 +40,0 @@

pyproject.toml

@@ -1,8 +0,10 @@
-[tool.poetry]
+[project]
 name = "mlserver"
-version = "1.7.1"
+version = "1.7.2-rc1"
 description = "MLServer"
-authors = ["Seldon Technologies Ltd. <[e-mail]>"]
+authors = [
+    {name = "Seldon Technologies Ltd.", email="[e-mail]"}]
 license = "Apache-2.0"
 readme = "README.md"
+requires-python = ">=3.9,<3.13"
 classifiers = [
@@ -13,4 +15,32 @@ "Operating System :: POSIX",
 include =  ["mlserver/rest/openapi/*.json"]
+dependencies = [
+    "click",
+    "fastapi>=0.88.0,!=0.89.0,<0.125.0",
+    "python-dotenv",
+    "grpcio>=1.67.1",
+    "numpy",
+    "pandas",
+    "protobuf>=5.29.5,<7.0.0",
+    "uvicorn>=0.38.0",
+    "starlette-exporter",
+    "py-grpc-prometheus",
+    "aiokafka",
+    "tritonclient[http]>=2.63",
+    "geventhttpclient",
+    "gevent",
+    "aiofiles",
+    "orjson>=3.10,<4",
+    "uvloop>=0.22.1,<0.24; sys_platform != 'win32' and (sys_platform != 'cygwin' and platform_python_implementation != 'PyPy')",
+    "pydantic>=2.7.1,<3.0.0",
+    "pydantic-settings>=2.3.0,<3.0.0",
+    "python-multipart",
+    # Required for Python <3.10 to access data files at runtime
+    "importlib-resources>=5.12,<7.0",
+    "opentelemetry-sdk>=1.22.0,<2.0.0",
+    "opentelemetry-instrumentation-fastapi>=0.43b0",
+    "opentelemetry-instrumentation-grpc>=0.43b0",
+    "opentelemetry-exporter-otlp-proto-grpc>=1.22.0,<2.0.0",
+]
 
-[tool.poetry.scripts]
+[project.scripts]
 mlserver = 'mlserver.cli:main'
@@ -45,38 +75,7 @@
 addopts = "--import-mode=importlib"
+norecursedirs = [
+    "runtimes"]
 
-[tool.poetry.dependencies]
-python = ">=3.9,<3.13"
-click = "*"
-fastapi = ">=0.88.0,!=0.89.0,<0.116.0"
-python-dotenv = "*"
-grpcio = ">=1.67.1"
-numpy = "*"
-pandas = "*"
-protobuf = ">=5.27.2,<7.0.0"
-uvicorn = "*"
-starlette-exporter = "*"
-py-grpc-prometheus = "*"
-aiokafka = "*"
-# add a min version to tritonclient due to https://github.com/triton-inference-server/server/issues/6246
-tritonclient = {version = ">=2.42", extras = ["http"]}
-geventhttpclient = "*"
-gevent = "*"
-aiofiles = "*"
-orjson = "*"
-uvloop = {version = "*", markers = "sys_platform != 'win32' and (sys_platform != 'cygwin' and platform_python_implementation != 'PyPy')"}
-pydantic = "^2.7.1"
-pydantic-settings = "^2.3.0"
-python-multipart = "*"
-
-## The importlib-resources backport is required to use some
-## functionality added in Python 3.10
-## https://setuptools.pypa.io/en/latest/userguide/datafiles.html#accessing-data-files-at-runtime
-importlib-resources = ">=5.12,<7.0"
-opentelemetry-sdk = "^1.22.0"
-opentelemetry-instrumentation-fastapi = ">=0.43b0"
-opentelemetry-instrumentation-grpc = ">=0.43b0"
-opentelemetry-exporter-otlp-proto-grpc = "^1.22.0"
-
 [tool.poetry.group.test.dependencies]
-tox = "4.16.0"
+tox = "4.24.2"
 
@@ -86,14 +85,14 @@ [tool.poetry.group.dev.dependencies]
 grpcio-tools = ">=1.67.1"
-pytest = "7.4.4"
-pytest-asyncio = "0.21.1"
-pytest-mock = "3.12.0"
-pytest-cases = "3.8.5"
-pytest-lazy-fixture = "^0.6.3"
-tox = "4.16.0"
+pytest = "8.4.2"
+exceptiongroup = "1.3.0"
+pytest-asyncio = "1.2.0"
+pytest-mock = "3.15.1"
+pytest-cases = "3.9.1"
+tox = "4.24.2"
 docker = "7.1.0"
-aiohttp = "3.12.8"
-aiohttp-retry = "2.8.3"
+aiohttp = "3.12.14"
+aiohttp-retry = "2.9.1"
 ## Used for FastAPI Async testing
 httpx = "0.27.0"
-kafka-python-ng = "2.2.3"
+kafka-python = "2.2.15"
 tenacity = "8.4.1"
@@ -103,5 +102,5 @@ pyyaml = "6.0.1"
 flake8 = "7.0.0"
-flake8-black = "0.3.6"
-mypy = "1.11.2"
-mypy-protobuf = "3.1.0"
+flake8-black = "0.4.0"
+mypy = "1.18.1"
+mypy-protobuf = "3.7.0"
 types-protobuf = "5.26.0.20240422"
@@ -111,3 +110,3 @@ types-orjson = "3.6.2"
 types-requests = "2.32.0.20250602"
-black = "24.8.0"
+black = "24.10.0"
 pip-licenses = "4.4.0"
@@ -134,2 +133,16 @@ pytest-xdist = "3.6.1"
 
+[tool.poetry.group.test-all-runtimes]
+optional = true
+
+[tool.poetry.group.test-all-runtimes.dependencies]
+mlserver-sklearn = {path = "./runtimes/sklearn", develop = false}
+mlserver-xgboost = {path = "./runtimes/xgboost", develop = false}
+mlserver-lightgbm = {path = "./runtimes/lightgbm", develop = false}
+mlserver-mlflow = {path = "./runtimes/mlflow", develop = false}
+mlserver-huggingface = {path = "./runtimes/huggingface", develop = false}
+mlserver-alibi-explain = {path = "./runtimes/alibi-explain", develop = false}
+mlserver-alibi-detect = {path = "./runtimes/alibi-detect", develop = false}
+mlserver-catboost = {path = "./runtimes/catboost", develop = false}
+
+
 [tool.poetry.group.all-runtimes-dev]
@@ -148,3 +161,3 @@ optional = true
 torchvision = "0.19.1"
-mlflow = "2.19.0"
+mlflow = ">=3.0.0"
 
@@ -151,0 +164,0 @@ ## Dev dependencies from HuggingFace

Alert delta unavailable

Currently unable to show alert delta for PyPI packages.

Improved metrics

Total package byte prevSize

483475

1.47%

Lines of code

11720

1.48%