onnx 1.16.1__cp310-cp310-win_amd64.whl → 1.17.0__cp310-cp310-win_amd64.whl

onnx/__init__.py

@@ -134,7 +134,9 @@ from onnx import (
 # Supported model formats that can be loaded from and saved to
 # The literals are formats with built-in support. But we also allow users to
 # register their own formats. So we allow str as well.
-_SupportedFormat = Union[Literal["protobuf", "textproto"], str]
+_SupportedFormat = Union[
+    Literal["protobuf", "textproto", "onnxtxt", "json"], str  # noqa: PYI051
+]
 # Default serialization format
 _DEFAULT_FORMAT = "protobuf"
 

onnx/_custom_element_types.py

@@ -0,0 +1,63 @@
+# Copyright (c) ONNX Project Contributors
+#
+# SPDX-License-Identifier: Apache-2.0
+"""This module defines custom dtypes not supported by numpy.
+Function :func:`onnx.numpy_helper.from_array`
+and :func:`onnx.numpy_helper.to_array` are using them
+to convert arrays from/to these types.
+Class :class:`onnx.reference.ReferenceEvalutor` also uses them.
+To create such an array for unit test for example, it is convenient to write
+something like the following:
+
+.. exec_code::
+
+    import numpy as np
+    from onnx import TensorProto
+    from onnx.reference.ops.op_cast import Cast_19 as Cast
+
+    tensor_bfloat16 = Cast.eval(np.array([0, 1], dtype=np.float32), to=TensorProto.BFLOAT16)
+
+The numpy representation dtypes used below are meant for internal use. They may change in the
+future based on the industry standardization of these numpy types.
+"""
+
+from __future__ import annotations
+
+import numpy as np
+
+import onnx
+
+#: Defines a bfloat16 as a uint16.
+bfloat16 = np.dtype((np.uint16, {"bfloat16": (np.uint16, 0)}))
+
+#: Defines float 8 e4m3fn type, see See :ref:`onnx-detail-float8` for technical details.
+float8e4m3fn = np.dtype((np.uint8, {"e4m3fn": (np.uint8, 0)}))
+
+#: Defines float 8 e4m3fnuz type, see See :ref:`onnx-detail-float8` for technical details.
+float8e4m3fnuz = np.dtype((np.uint8, {"e4m3fnuz": (np.uint8, 0)}))
+
+#: Defines float 8 e5m2 type, see See :ref:`onnx-detail-float8` for technical details.
+float8e5m2 = np.dtype((np.uint8, {"e5m2": (np.uint8, 0)}))
+
+#: Defines float 8 e5m2fnuz type, see See :ref:`onnx-detail-float8` for technical details.
+float8e5m2fnuz = np.dtype((np.uint8, {"e5m2fnuz": (np.uint8, 0)}))
+
+#: Defines int4, see See :ref:`onnx-detail-int4` for technical details.
+#: Do note that one integer is stored using a byte and therefore is twice bigger
+#: than its onnx size.
+uint4 = np.dtype((np.uint8, {"uint4": (np.uint8, 0)}))
+
+#: Defines int4, see See :ref:`onnx-detail-int4` for technical details.
+#: Do note that one integer is stored using a byte and therefore is twice bigger
+#: than its onnx size.
+int4 = np.dtype((np.int8, {"int4": (np.int8, 0)}))
+
+mapping_name_to_data_type = {
+    "bfloat16": onnx.TensorProto.BFLOAT16,
+    "e4m3fn": onnx.TensorProto.FLOAT8E4M3FN,
+    "e4m3fnuz": onnx.TensorProto.FLOAT8E4M3FNUZ,
+    "e5m2": onnx.TensorProto.FLOAT8E5M2,
+    "e5m2fnuz": onnx.TensorProto.FLOAT8E5M2FNUZ,
+    "int4": onnx.TensorProto.INT4,
+    "uint4": onnx.TensorProto.UINT4,
+}

onnx/backend/base.py

@@ -1,10 +1,10 @@
 # Copyright (c) ONNX Project Contributors
 #
 # SPDX-License-Identifier: Apache-2.0
-
+from __future__ import annotations
 
 from collections import namedtuple
-from typing import Any, Dict, NewType, Optional, Sequence, Tuple, Type
+from typing import Any, NewType, Sequence
 
 import numpy
 
@@ -37,11 +37,11 @@ class Device:
 
 def namedtupledict(
     typename: str, field_names: Sequence[str], *args: Any, **kwargs: Any
-) -> Type[Tuple[Any, ...]]:
+) -> type[tuple[Any, ...]]:
     field_names_map = {n: i for i, n in enumerate(field_names)}
     # Some output names are invalid python identifier, e.g. "0"
     kwargs.setdefault("rename", True)
-    data = namedtuple(typename, field_names, *args, **kwargs)  # type: ignore
+    data = namedtuple(typename, field_names, *args, **kwargs)  # type: ignore  # noqa: PYI024
 
     def getitem(self: Any, key: Any) -> Any:
         if isinstance(key, str):
@@ -58,7 +58,7 @@ class BackendRep:
     BackendRep to retrieve the corresponding results.
     """
 
-    def run(self, inputs: Any, **kwargs: Any) -> Tuple[Any, ...]:
+    def run(self, inputs: Any, **kwargs: Any) -> tuple[Any, ...]:  # noqa: ARG002
         """Abstract function."""
         return (None,)
 
@@ -76,15 +76,15 @@ class Backend:
 
     @classmethod
     def is_compatible(
-        cls, model: ModelProto, device: str = "CPU", **kwargs: Any
+        cls, model: ModelProto, device: str = "CPU", **kwargs: Any  # noqa: ARG003
     ) -> bool:
         # Return whether the model is compatible with the backend.
         return True
 
     @classmethod
     def prepare(
-        cls, model: ModelProto, device: str = "CPU", **kwargs: Any
-    ) -> Optional[BackendRep]:
+        cls, model: ModelProto, device: str = "CPU", **kwargs: Any  # noqa: ARG003
+    ) -> BackendRep | None:
         # TODO Remove Optional from return type
         onnx.checker.check_model(model)
         return None
@@ -92,7 +92,7 @@ class Backend:
     @classmethod
     def run_model(
         cls, model: ModelProto, inputs: Any, device: str = "CPU", **kwargs: Any
-    ) -> Tuple[Any, ...]:
+    ) -> tuple[Any, ...]:
         backend = cls.prepare(model, device, **kwargs)
         assert backend is not None
         return backend.run(inputs)
@@ -101,11 +101,13 @@ class Backend:
     def run_node(
         cls,
         node: NodeProto,
-        inputs: Any,
-        device: str = "CPU",
-        outputs_info: Optional[Sequence[Tuple[numpy.dtype, Tuple[int, ...]]]] = None,
-        **kwargs: Dict[str, Any],
-    ) -> Optional[Tuple[Any, ...]]:
+        inputs: Any,  # noqa: ARG003
+        device: str = "CPU",  # noqa: ARG003
+        outputs_info: (  # noqa: ARG003
+            Sequence[tuple[numpy.dtype, tuple[int, ...]]] | None
+        ) = None,
+        **kwargs: dict[str, Any],
+    ) -> tuple[Any, ...] | None:
         """Simple run one operator and return the results.
 
         Args:
@@ -130,7 +132,7 @@ class Backend:
         return None
 
     @classmethod
-    def supports_device(cls, device: str) -> bool:
+    def supports_device(cls, device: str) -> bool:  # noqa: ARG003
         """Checks whether the backend is compiled with particular device support.
         In particular it's used in the testing suite.
         """

onnx/backend/sample/ops/__init__.py

@@ -1,22 +1,22 @@
 # Copyright (c) ONNX Project Contributors
 
 # SPDX-License-Identifier: Apache-2.0
+from __future__ import annotations
 
 import importlib
 import inspect
 import pkgutil
 import sys
 from types import ModuleType
-from typing import Dict
 
 
-def collect_sample_implementations() -> Dict[str, str]:
-    dict_: Dict[str, str] = {}
+def collect_sample_implementations() -> dict[str, str]:
+    dict_: dict[str, str] = {}
     _recursive_scan(sys.modules[__name__], dict_)
     return dict_
 
 
-def _recursive_scan(package: ModuleType, dict_: Dict[str, str]) -> None:
+def _recursive_scan(package: ModuleType, dict_: dict[str, str]) -> None:
     pkg_dir = package.__path__  # type: ignore
     module_location = package.__name__
     for _module_loader, name, ispkg in pkgutil.iter_modules(pkg_dir):  # type: ignore

onnx/backend/sample/ops/abs.py

@@ -1,4 +1,5 @@
 # SPDX-License-Identifier: Apache-2.0
+from __future__ import annotations
 
 import numpy as np
 

onnx/backend/test/__init__.py

@@ -1,6 +1,7 @@
 # Copyright (c) ONNX Project Contributors
 #
 # SPDX-License-Identifier: Apache-2.0
+from __future__ import annotations
 
 __all__ = ["BackendTest"]
 # for backward compatibility

onnx/backend/test/case/__init__.py

@@ -1,14 +1,14 @@
 # Copyright (c) ONNX Project Contributors
 
 # SPDX-License-Identifier: Apache-2.0
+from __future__ import annotations
 
 import sys
-from typing import Dict, List, Tuple
 
 from onnx.backend.test.case.base import Snippets
 from onnx.backend.test.case.utils import import_recursive
 
 
-def collect_snippets() -> Dict[str, List[Tuple[str, str]]]:
+def collect_snippets() -> dict[str, list[tuple[str, str]]]:
     import_recursive(sys.modules[__name__])
     return Snippets

onnx/backend/test/case/base.py

@@ -1,16 +1,17 @@
 # Copyright (c) ONNX Project Contributors
 
 # SPDX-License-Identifier: Apache-2.0
+from __future__ import annotations
 
 import inspect
 from collections import defaultdict
 from textwrap import dedent
-from typing import Any, ClassVar, Dict, List, Tuple, Type
+from typing import Any, ClassVar
 
 import numpy as np
 
 
-def process_snippet(op_name: str, name: str, export: Any) -> Tuple[str, str]:
+def process_snippet(op_name: str, name: str, export: Any) -> tuple[str, str]:
     snippet_name = name[len("export_") :] or op_name.lower()
     source_code = dedent(inspect.getsource(export))
     # remove the function signature line
@@ -20,14 +21,14 @@ def process_snippet(op_name: str, name: str, export: Any) -> Tuple[str, str]:
     return snippet_name, dedent("\n".join(lines[2:]))
 
 
-Snippets: Dict[str, List[Tuple[str, str]]] = defaultdict(list)
+Snippets: dict[str, list[tuple[str, str]]] = defaultdict(list)
 
 
 class _Exporter(type):
-    exports: ClassVar[Dict[str, List[Tuple[str, str]]]] = defaultdict(list)
+    exports: ClassVar[dict[str, list[tuple[str, str]]]] = defaultdict(list)
 
     def __init__(
-        cls, name: str, bases: Tuple[Type[Any], ...], dct: Dict[str, Any]
+        cls, name: str, bases: tuple[type[Any], ...], dct: dict[str, Any]
     ) -> None:
         for k, v in dct.items():
             if k.startswith("export"):

onnx/backend/test/case/model/__init__.py

@@ -1,9 +1,10 @@
 # Copyright (c) ONNX Project Contributors
 
 # SPDX-License-Identifier: Apache-2.0
+from __future__ import annotations
 
 import sys
-from typing import List, Optional, Sequence
+from typing import Sequence
 
 import numpy as np
 
@@ -18,7 +19,7 @@ def expect(
     model: ModelProto,
     inputs: Sequence[np.ndarray],
     outputs: Sequence[np.ndarray],
-    name: Optional[str] = None,
+    name: str | None = None,
 ) -> None:
     name = name or model.graph.name
     _SimpleModelTestCases.append(
@@ -40,7 +41,7 @@ def expect(
 BASE_URL = "onnx/backend/test/data/light/light_%s.onnx"
 
 
-def collect_testcases() -> List[TestCase]:
+def collect_testcases() -> list[TestCase]:
     """Collect model test cases defined in python/numpy code."""
     real_model_testcases = []
 

onnx/backend/test/case/model/expand.py

@@ -1,6 +1,7 @@
 # Copyright (c) ONNX Project Contributors
 
 # SPDX-License-Identifier: Apache-2.0
+from __future__ import annotations
 
 from typing import Sequence
 

onnx/backend/test/case/model/gradient.py

@@ -1,6 +1,7 @@
 # Copyright (c) ONNX Project Contributors
 
 # SPDX-License-Identifier: Apache-2.0
+from __future__ import annotations
 
 import numpy as np
 

onnx/backend/test/case/model/sequence.py

@@ -64,7 +64,9 @@ def SplitToSequenceImpl(
             results = np.array_split(tensor, split_indices, axis)
             return [np.squeeze(res, axis) for res in results]
     if np.isscalar(split):
-        split_indices = [i * split + 1 for i in range(dim_size) if i * split + 1 < dim_size]  # type: ignore
+        split_indices = [
+            i * split + 1 for i in range(dim_size) if i * split + 1 < dim_size
+        ]  # type: ignore
     else:
         split_indices = np.cumsum(split) + 1
     return np.array_split(tensor, split_indices, axis)  # type: ignore

onnx/backend/test/case/model/shrink.py

@@ -1,6 +1,7 @@
 # Copyright (c) ONNX Project Contributors
 
 # SPDX-License-Identifier: Apache-2.0
+from __future__ import annotations
 
 import numpy as np
 

onnx/backend/test/case/model/sign.py

@@ -1,6 +1,7 @@
 # Copyright (c) ONNX Project Contributors
 
 # SPDX-License-Identifier: Apache-2.0
+from __future__ import annotations
 
 import numpy as np
 

onnx/backend/test/case/model/single-relu.py

@@ -1,6 +1,7 @@
 # Copyright (c) ONNX Project Contributors
 
 # SPDX-License-Identifier: Apache-2.0
+from __future__ import annotations
 
 import numpy as np
 

onnx/backend/test/case/model/stringnormalizer.py

@@ -1,7 +1,7 @@
 # Copyright (c) ONNX Project Contributors
 
 # SPDX-License-Identifier: Apache-2.0
-
+from __future__ import annotations
 
 from typing import Sequence
 

onnx/backend/test/case/node/__init__.py

@@ -1,12 +1,13 @@
 # Copyright (c) ONNX Project Contributors
 #
 # SPDX-License-Identifier: Apache-2.0
+from __future__ import annotations
 
 import subprocess
 import sys
 from copy import deepcopy
 from pathlib import Path
-from typing import Any, Callable, Dict, List, Optional, Sequence, Tuple, Union
+from typing import Any, Callable, Sequence
 
 import numpy as np
 
@@ -26,12 +27,13 @@ from onnx.onnx_pb import (
 _NodeTestCases = []
 _TargetOpType = None
 _DiffOpTypes = None
+_existing_names: dict[str, onnx.NodeProto] = {}
 
 
 def _rename_edges_helper(
     internal_node: NodeProto,
     rename_helper: Callable[[str], str],
-    attribute_map: Dict[str, AttributeProto],
+    attribute_map: dict[str, AttributeProto],
     prefix: str,
 ) -> NodeProto:
     new_node = NodeProto()
@@ -63,17 +65,13 @@ def _rename_edges_helper(
                 for init_desc in new_graph.initializer:
                     sg_rename[init_desc.name] = init_desc.name = prefix + init_desc.name
                 for sparse_init_desc in new_graph.sparse_initializer:
-                    sg_rename[
+                    sg_rename[sparse_init_desc.values.name] = (
                         sparse_init_desc.values.name
-                    ] = sparse_init_desc.values.name = (
-                        prefix + sparse_init_desc.values.name
-                    )
+                    ) = (prefix + sparse_init_desc.values.name)
                 for sparse_init_desc in new_graph.sparse_initializer:
-                    sg_rename[
+                    sg_rename[sparse_init_desc.indices.name] = (
                         sparse_init_desc.indices.name
-                    ] = sparse_init_desc.indices.name = (
-                        prefix + sparse_init_desc.indices.name
-                    )
+                    ) = (prefix + sparse_init_desc.indices.name)
 
                 def subgraph_rename_helper(name: str) -> Any:
                     if name in sg_rename:  # noqa: B023
@@ -95,7 +93,7 @@ def _rename_edges_helper(
 # FIXME(TMVector): Any reason we can't get rid of this and use the C++ helper directly?
 def function_expand_helper(
     node: NodeProto, function_proto: FunctionProto, op_prefix: str
-) -> List[NodeProto]:
+) -> list[NodeProto]:
     io_names_map = {}
     attribute_map = {a.name: a for a in node.attribute}
 
@@ -129,8 +127,8 @@ def function_expand_helper(
 
 
 def function_testcase_helper(
-    node: NodeProto, input_types: List[TypeProto], name: str
-) -> Tuple[List[Tuple[List[NodeProto], Any]], int]:
+    node: NodeProto, input_types: list[TypeProto], name: str
+) -> tuple[list[tuple[list[NodeProto], Any]], int]:
     test_op = node.op_type
     op_prefix = test_op + "_" + name + "_expanded_function_"
     schema = onnx.defs.get_schema(test_op, domain=node.domain)
@@ -170,9 +168,9 @@ def function_testcase_helper(
 
 
 def _extract_value_info(
-    input: Union[List[Any], np.ndarray, None],
+    input: list[Any] | np.ndarray | None,
     name: str,
-    type_proto: Optional[TypeProto] = None,
+    type_proto: TypeProto | None = None,
 ) -> onnx.ValueInfoProto:
     if type_proto is None:
         if input is None:
@@ -197,7 +195,13 @@ def _extract_value_info(
 
 
 def _make_test_model_gen_version(graph: GraphProto, **kwargs: Any) -> ModelProto:
-    latest_onnx_version, latest_ml_version, latest_training_version = onnx.helper.VERSION_TABLE[-1][2:5]  # type: ignore
+    (
+        latest_onnx_version,
+        latest_ml_version,
+        latest_training_version,
+    ) = onnx.helper.VERSION_TABLE[-1][
+        2:5
+    ]  # type: ignore
     if "opset_imports" in kwargs:
         for opset in kwargs["opset_imports"]:
             # If the test model uses an unreleased opset version (latest_version+1),
@@ -236,8 +240,8 @@ def _make_test_model_gen_version(graph: GraphProto, **kwargs: Any) -> ModelProto
 # the latest opset vesion that supports before targeted opset version
 def expect(
     node_op: onnx.NodeProto,
-    inputs: Sequence[Union[np.ndarray, TensorProto]],
-    outputs: Sequence[Union[np.ndarray, TensorProto]],
+    inputs: Sequence[np.ndarray | TensorProto],
+    outputs: Sequence[np.ndarray | TensorProto],
     name: str,
     **kwargs: Any,
 ) -> None:
@@ -246,6 +250,11 @@ def expect(
         return
     if _DiffOpTypes is not None and node_op.op_type.lower() not in _DiffOpTypes:
         return
+    if name in _existing_names:
+        raise ValueError(
+            f"Name {name!r} is already using by one test case for node type {node_op.op_type!r}."
+        )
+    _existing_names[name] = node_op
 
     # in case node_op is modified
     node = deepcopy(node_op)
@@ -303,8 +312,8 @@ def expect(
     # Create list of types for node.input, filling a default TypeProto for missing inputs:
     # E.g. merge(["x", "", "y"], [x-value-info, y-value-info]) will return [x-type, default-type, y-type]
     def merge(
-        node_inputs: List[str], present_value_info: List[onnx.ValueInfoProto]
-    ) -> List[TypeProto]:
+        node_inputs: list[str], present_value_info: list[onnx.ValueInfoProto]
+    ) -> list[TypeProto]:
         if node_inputs:
             if node_inputs[0] != "":
                 return [
@@ -372,7 +381,7 @@ def expect(
         )
 
 
-def collect_testcases(op_type: str) -> List[TestCase]:
+def collect_testcases(op_type: str) -> list[TestCase]:
     """Collect node test cases"""
     # only keep those tests related to this operator
     global _TargetOpType  # noqa: PLW0603
@@ -382,7 +391,7 @@ def collect_testcases(op_type: str) -> List[TestCase]:
     return _NodeTestCases
 
 
-def collect_diff_testcases() -> List[TestCase]:
+def collect_diff_testcases() -> list[TestCase]:
     """Collect node test cases which are different from the main branch"""
     global _DiffOpTypes  # noqa: PLW0603
     _DiffOpTypes = get_diff_op_types()

onnx/backend/test/case/node/_image_decoder_data.py

@@ -6,6 +6,7 @@
 # They were generated by image_decoder.py's generate_test_data with pillow
 # To make ONNX backend test generation simple, use freeze NumPy array directly
 # People can still use generate_test_data to get these data here with installed pillow
+from __future__ import annotations
 
 from typing import NamedTuple
 

onnx/backend/test/case/node/abs.py

@@ -1,6 +1,7 @@
 # Copyright (c) ONNX Project Contributors
 #
 # SPDX-License-Identifier: Apache-2.0
+from __future__ import annotations
 
 import numpy as np
 

onnx/backend/test/case/node/acos.py

@@ -1,6 +1,7 @@
 # Copyright (c) ONNX Project Contributors
 #
 # SPDX-License-Identifier: Apache-2.0
+from __future__ import annotations
 
 import numpy as np
 

onnx/backend/test/case/node/acosh.py

@@ -1,6 +1,7 @@
 # Copyright (c) ONNX Project Contributors
 #
 # SPDX-License-Identifier: Apache-2.0
+from __future__ import annotations
 
 import numpy as np
 

onnx/backend/test/case/node/adagrad.py

@@ -1,6 +1,7 @@
 # Copyright (c) ONNX Project Contributors
 #
 # SPDX-License-Identifier: Apache-2.0
+from __future__ import annotations
 
 import numpy as np
 
@@ -21,7 +22,7 @@ def apply_adagrad(r, t, x, g, h, norm_coefficient, epsilon, decay_factor):  # ty
     h_sqrt = np.sqrt(h_new) + epsilon
     # Apply ADAGRAD update rule.
     x_new = x - r_ * g_regularized / h_sqrt
-    return (x_new, h_new)
+    return (x_new.astype(x.dtype), h_new.astype(h.dtype))
 
 
 class Adagrad(Base):

onnx/backend/test/case/node/adam.py

@@ -1,6 +1,7 @@
 # Copyright (c) ONNX Project Contributors
 #
 # SPDX-License-Identifier: Apache-2.0
+from __future__ import annotations
 
 import numpy as np
 
@@ -10,7 +11,9 @@ from onnx.backend.test.case.node import expect
 from onnx.defs import AI_ONNX_PREVIEW_TRAINING_DOMAIN
 
 
-def apply_adam(r, t, x, g, v, h, norm_coefficient, norm_coefficient_post, alpha, beta, epsilon):  # type: ignore
+def apply_adam(
+    r, t, x, g, v, h, norm_coefficient, norm_coefficient_post, alpha, beta, epsilon
+):  # type: ignore
     # Add gradient of regularization term.
     g_regularized = norm_coefficient * x + g
     # Update momentum.

onnx/backend/test/case/node/add.py

@@ -1,6 +1,7 @@
 # Copyright (c) ONNX Project Contributors
 #
 # SPDX-License-Identifier: Apache-2.0
+from __future__ import annotations
 
 import numpy as np
 

onnx/backend/test/case/node/affinegrid.py

@@ -1,6 +1,7 @@
 # Copyright (c) ONNX Project Contributors
 #
 # SPDX-License-Identifier: Apache-2.0
+from __future__ import annotations
 
 import numpy as np
 

onnx/backend/test/case/node/ai_onnx_ml/array_feature_extractor.py

@@ -1,6 +1,7 @@
 # Copyright (c) ONNX Project Contributors
 
 # SPDX-License-Identifier: Apache-2.0
+from __future__ import annotations
 
 import numpy as np
 

onnx/backend/test/case/node/ai_onnx_ml/binarizer.py

@@ -1,6 +1,7 @@
 # Copyright (c) ONNX Project Contributors
 
 # SPDX-License-Identifier: Apache-2.0
+from __future__ import annotations
 
 import numpy as np
 

onnx/backend/test/case/node/ai_onnx_ml/label_encoder.py

@@ -1,6 +1,7 @@
 # Copyright (c) ONNX Project Contributors
 
 # SPDX-License-Identifier: Apache-2.0
+from __future__ import annotations
 
 import numpy as np
 

onnx/backend/test/case/node/ai_onnx_ml/tree_ensemble.py

@@ -1,6 +1,7 @@
 # Copyright (c) ONNX Project Contributors
 
 # SPDX-License-Identifier: Apache-2.0
+from __future__ import annotations
 
 import numpy as np
 

onnx/backend/test/case/node/and.py

@@ -1,6 +1,7 @@
 # Copyright (c) ONNX Project Contributors
 #
 # SPDX-License-Identifier: Apache-2.0
+from __future__ import annotations
 
 import numpy as np
 

onnx/backend/test/case/node/argmax.py

@@ -1,6 +1,7 @@
 # Copyright (c) ONNX Project Contributors
 #
 # SPDX-License-Identifier: Apache-2.0
+from __future__ import annotations
 
 import numpy as np
 

onnx/backend/test/case/node/argmin.py

@@ -1,6 +1,7 @@
 # Copyright (c) ONNX Project Contributors
 #
 # SPDX-License-Identifier: Apache-2.0
+from __future__ import annotations
 
 import numpy as np