onnx 1.16.2__cp311-cp311-win32.whl → 1.18.0__cp311-cp311-win32.whl

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

@@ -1,6 +1,7 @@
 # Copyright (c) ONNX Project Contributors
 #
 # SPDX-License-Identifier: Apache-2.0
+from __future__ import annotations
 
 import numpy as np
 
@@ -286,7 +287,16 @@ class AveragePool(Base):
             constant_values=np.nan,
         )
         pads = (pad_top, pad_left, pad_bottom, pad_right)
-        y = pool(padded, x_shape, kernel_shape, strides, out_shape, "AVG", pads)
+        y = pool(
+            padded,
+            x_shape,
+            kernel_shape,
+            strides,
+            out_shape,
+            "AVG",
+            pads_required=pads,
+            pads=pads,
+        )
 
         expect(node, inputs=[x], outputs=[y], name="test_averagepool_2d_same_upper")
 
@@ -324,7 +334,16 @@ class AveragePool(Base):
             constant_values=np.nan,
         )
         pads = (pad_top, pad_left, pad_bottom, pad_right)
-        y = pool(padded, x_shape, kernel_shape, strides, out_shape, "AVG", pads)
+        y = pool(
+            padded,
+            x_shape,
+            kernel_shape,
+            strides,
+            out_shape,
+            "AVG",
+            pads_required=pads,
+            pads=pads,
+        )
 
         expect(node, inputs=[x], outputs=[y], name="test_averagepool_2d_same_lower")
 
@@ -350,16 +369,30 @@ class AveragePool(Base):
         pad_right = 2
         pad_left = 2
         pads = [pad_top, pad_left, pad_bottom, pad_right]
-        out_shape, pads = get_output_shape_explicit_padding(
+        out_shape, extra_pads = get_output_shape_explicit_padding(
             pads, x_shape[2:], kernel_shape, strides, ceil_mode=False
         )
         padded = np.pad(
             x,
-            ((0, 0), (0, 0), (pads[0], pads[2]), (pads[1], pads[3])),
+            (
+                (0, 0),
+                (0, 0),
+                (extra_pads[0], extra_pads[2]),
+                (extra_pads[1], extra_pads[3]),
+            ),
             mode="constant",
             constant_values=np.nan,
         )
-        y = pool(padded, x_shape, kernel_shape, strides, out_shape, "AVG", pads)
+        y = pool(
+            padded,
+            x_shape,
+            kernel_shape,
+            strides,
+            out_shape,
+            "AVG",
+            pads_required=extra_pads,
+            pads=pads,
+        )
 
         expect(node, inputs=[x], outputs=[y], name="test_averagepool_2d_pads")
 
@@ -387,12 +420,17 @@ class AveragePool(Base):
         pad_right = 2
         pad_left = 2
         pads = [pad_top, pad_left, pad_bottom, pad_right]
-        out_shape, pads = get_output_shape_explicit_padding(
+        out_shape, extra_pads = get_output_shape_explicit_padding(
             pads, x_shape[2:], kernel_shape, strides, dilations, ceil_mode=False
         )
         padded = np.pad(
             x,
-            ((0, 0), (0, 0), (pads[0], pads[2]), (pads[1], pads[3])),
+            (
+                (0, 0),
+                (0, 0),
+                (extra_pads[0], extra_pads[2]),
+                (extra_pads[1], extra_pads[3]),
+            ),
             mode="constant",
             constant_values=0,
         )
@@ -403,7 +441,8 @@ class AveragePool(Base):
             strides,
             out_shape,
             "AVG",
-            pads,
+            pads_required=extra_pads,
+            pads=pads,
             count_include_pad=1,
         )
 
@@ -434,7 +473,16 @@ class AveragePool(Base):
             None, x_shape[2:], kernel_shape, strides, ceil_mode=False
         )
         padded = x
-        y = pool(padded, x_shape, kernel_shape, strides, out_shape, "AVG", pads)
+        y = pool(
+            padded,
+            x_shape,
+            kernel_shape,
+            strides,
+            out_shape,
+            "AVG",
+            pads_required=pads,
+            pads=None,
+        )
 
         expect(node, inputs=[x], outputs=[y], name="test_averagepool_2d_strides")
 
@@ -467,6 +515,39 @@ class AveragePool(Base):
 
         expect(node, inputs=[x], outputs=[y], name="test_averagepool_2d_ceil")
 
+    @staticmethod
+    def export_averagepool_2d_ceil_last_window_starts_on_pad() -> None:
+        """input_shape: [1, 3, 2, 2]
+        output_shape: [1, 3, 1, 1]
+        """
+        node = onnx.helper.make_node(
+            "AveragePool",
+            inputs=["x"],
+            outputs=["y"],
+            kernel_shape=[3, 3],
+            strides=[3, 3],
+            pads=[1, 1, 1, 1],
+            ceil_mode=True,
+            count_include_pad=1,
+        )
+        x = np.array(
+            [
+                [
+                    [[0.8580, 0.0786], [0.2692, 0.1537]],
+                    [[0.8816, 0.4353], [0.5772, 0.6623]],
+                    [[0.9067, 0.9483], [0.5970, 0.7630]],
+                ]
+            ]
+        ).astype(np.float32)
+        y = np.array([[[[0.1511]], [[0.2841]], [[0.3572]]]]).astype(np.float32)
+
+        expect(
+            node,
+            inputs=[x],
+            outputs=[y],
+            name="test_averagepool_2d_ceil_last_window_starts_on_pad",
+        )
+
     @staticmethod
     def export_averagepool_2d_dilations() -> None:
         """input_shape: [1, 1, 4, 4]
@@ -577,7 +658,7 @@ class AveragePool(Base):
                 )
 
                 x = np.random.randn(1, 1, *x_shape).astype(np.float32)
-                out_shape, pads = get_output_shape_explicit_padding(
+                out_shape, extra_pads = get_output_shape_explicit_padding(
                     None,
                     x_shape,
                     kernel_shape,
@@ -590,9 +671,9 @@ class AveragePool(Base):
                     (
                         (0, 0),
                         (0, 0),
-                        (pads[0], pads[3]),
-                        (pads[1], pads[4]),
-                        (pads[2], pads[5]),
+                        (extra_pads[0], extra_pads[3]),
+                        (extra_pads[1], extra_pads[4]),
+                        (extra_pads[2], extra_pads[5]),
                     ),
                     mode="constant",
                     constant_values=0 if count_include_pad == 1 else np.nan,
@@ -604,7 +685,8 @@ class AveragePool(Base):
                     strides,
                     out_shape,
                     "AVG",
-                    pads=pads,
+                    pads_required=extra_pads,
+                    pads=None,
                     dilations=dilations,
                     count_include_pad=count_include_pad,
                 )

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

@@ -1,6 +1,7 @@
 # Copyright (c) ONNX Project Contributors
 #
 # SPDX-License-Identifier: Apache-2.0
+from __future__ import annotations
 
 import numpy as np
 
@@ -9,7 +10,7 @@ from onnx.backend.test.case.base import Base
 from onnx.backend.test.case.node import expect
 
 
-def _batchnorm_test_mode(x, s, bias, mean, var, epsilon=1e-5):  # type: ignore
+def _batchnorm_test_mode(x, s, bias, mean, var, epsilon=1e-5):
     dims_x = len(x.shape)
     dim_ones = (1,) * (dims_x - 2)
     s = s.reshape(-1, *dim_ones)
@@ -19,7 +20,7 @@ def _batchnorm_test_mode(x, s, bias, mean, var, epsilon=1e-5):  # type: ignore
     return s * (x - mean) / np.sqrt(var + epsilon) + bias
 
 
-def _batchnorm_training_mode(x, s, bias, mean, var, momentum=0.9, epsilon=1e-5):  # type: ignore
+def _batchnorm_training_mode(x, s, bias, mean, var, momentum=0.9, epsilon=1e-5):
     axis = tuple(np.delete(np.arange(len(x.shape)), 1))
     saved_mean = x.mean(axis=axis)
     saved_var = x.var(axis=axis)

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

@@ -1,6 +1,7 @@
 # Copyright (c) ONNX Project Contributors
 #
 # SPDX-License-Identifier: Apache-2.0
+from __future__ import annotations
 
 import numpy as np
 
@@ -9,7 +10,7 @@ from onnx.backend.test.case.base import Base
 from onnx.backend.test.case.node import expect
 
 
-def bernoulli_reference_implementation(x, dtype):  # type: ignore
+def bernoulli_reference_implementation(x, dtype):
     # binomial n = 1 equal bernoulli
     # This example and test-case is for informational purpose. The generator operator is
     # non-deterministic and may not produce the same values in different implementations

onnx/backend/test/case/node/bitshift.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/bitwiseand.py

@@ -1,8 +1,9 @@
 # Copyright (c) ONNX Project Contributors
 #
 # SPDX-License-Identifier: Apache-2.0
+from __future__ import annotations
 
-import numpy as np  # type: ignore
+import numpy as np
 
 import onnx
 from onnx.backend.test.case.base import Base

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

@@ -1,8 +1,9 @@
 # Copyright (c) ONNX Project Contributors
 #
 # SPDX-License-Identifier: Apache-2.0
+from __future__ import annotations
 
-import numpy as np  # type: ignore
+import numpy as np
 
 import onnx
 from onnx.backend.test.case.base import Base

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

@@ -1,8 +1,9 @@
 # Copyright (c) ONNX Project Contributors
 #
 # SPDX-License-Identifier: Apache-2.0
+from __future__ import annotations
 
-import numpy as np  # type: ignore
+import numpy as np
 
 import onnx
 from onnx.backend.test.case.base import Base

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

@@ -1,8 +1,9 @@
 # Copyright (c) ONNX Project Contributors
 #
 # SPDX-License-Identifier: Apache-2.0
+from __future__ import annotations
 
-import numpy as np  # type: ignore
+import numpy as np
 
 import onnx
 from onnx.backend.test.case.base import Base

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

@@ -1,7 +1,7 @@
 # Copyright (c) ONNX Project Contributors
 #
 # SPDX-License-Identifier: Apache-2.0
-
+from __future__ import annotations
 
 import numpy as np
 
@@ -26,7 +26,12 @@ class BlackmanWindow(Base):
         y = a0
         y += a1 * np.cos(2 * np.pi * np.arange(0, size, 1, dtype=np.float32) / size)
         y += a2 * np.cos(4 * np.pi * np.arange(0, size, 1, dtype=np.float32) / size)
-        expect(node, inputs=[size], outputs=[y], name="test_blackmanwindow")
+        expect(
+            node,
+            inputs=[size],
+            outputs=[y.astype(np.float32)],
+            name="test_blackmanwindow",
+        )
 
         # Test symmetric window
         node = onnx.helper.make_node(
@@ -43,4 +48,9 @@ class BlackmanWindow(Base):
         y += a2 * np.cos(
             4 * np.pi * np.arange(0, size, 1, dtype=np.float32) / (size - 1)
         )
-        expect(node, inputs=[size], outputs=[y], name="test_blackmanwindow_symmetric")
+        expect(
+            node,
+            inputs=[size],
+            outputs=[y.astype(np.float32)],
+            name="test_blackmanwindow_symmetric",
+        )

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

@@ -1,13 +1,14 @@
 # Copyright (c) ONNX Project Contributors
 #
 # SPDX-License-Identifier: Apache-2.0
+from __future__ import annotations
 
 import sys
 
 import numpy as np
 
 import onnx
-import onnx.reference.custom_element_types as custom
+import onnx._custom_element_types as custom
 from onnx import TensorProto, helper, subbyte
 from onnx.backend.test.case.base import Base
 from onnx.backend.test.case.node import expect
@@ -17,7 +18,11 @@ from onnx.helper import (
     make_tensor,
     tensor_dtype_to_field,
 )
-from onnx.numpy_helper import float8e4m3_to_float32, float8e5m2_to_float32
+from onnx.numpy_helper import (
+    float8e4m3_to_float32,
+    float8e5m2_to_float32,
+    unpacked_float4e2m1_to_float32,
+)
 
 
 class Cast(Base):
@@ -61,6 +66,10 @@ class Cast(Base):
             ("INT4", "FLOAT"),
             ("INT4", "FLOAT16"),
             ("INT4", "INT8"),
+            ("FLOAT4E2M1", "FLOAT"),
+            ("FLOAT4E2M1", "FLOAT16"),
+            ("FLOAT", "FLOAT4E2M1"),
+            ("FLOAT16", "FLOAT4E2M1"),
         ]
 
         vect_float32_to_float8e4m3 = np.vectorize(float32_to_float8e4m3)
@@ -277,7 +286,57 @@ class Cast(Base):
                 output_type_proto = onnx.helper.make_tensor_type_proto(
                     getattr(TensorProto, to_type), input_shape
                 )
+            elif from_type == "FLOAT4E2M1" or to_type == "FLOAT4E2M1":
+                np_fp32 = np.array(
+                    [
+                        "0.48",
+                        "0.25",
+                        "1.05",
+                        "-3.5",
+                        "-8",
+                        "9",
+                        "1000000",
+                        "1e-7",
+                        "NaN",
+                        "INF",
+                        "+INF",
+                        "-INF",
+                        "-4",
+                        "0.01",
+                        "-0.0",
+                    ],
+                    dtype=np.float32,
+                )
+                input_shape = (3, 5)
+                if from_type == "FLOAT":
+                    input_values = np_fp32
+                    input = make_tensor(
+                        "x", TensorProto.FLOAT, input_shape, input_values.tolist()
+                    )
+                elif from_type == "FLOAT16":
+                    input_values = np_fp32.astype(np.float16).astype(np.float32)
+                    input = make_tensor(
+                        "x", TensorProto.FLOAT16, input_shape, input_values.tolist()
+                    )
+                elif from_type == "FLOAT4E2M1":
+                    input = make_tensor(
+                        "x", TensorProto.FLOAT4E2M1, input_shape, np_fp32.tolist()
+                    )
+                else:
+                    raise ValueError(
+                        f"Conversion from {from_type} to {to_type} is not tested."
+                    )
 
+                if to_type not in ("FLOAT", "FLOAT16", "FLOAT4E2M1"):
+                    raise ValueError(
+                        f"Conversion from {from_type} to {to_type} is not tested."
+                    )
+                expected = unpacked_float4e2m1_to_float32(
+                    subbyte.float32_to_float4e2m1_unpacked(np_fp32)
+                )
+                output = make_tensor(
+                    "y", getattr(TensorProto, to_type), input_shape, expected.tolist()
+                )
             elif from_type != "STRING":
                 input = np.random.random_sample(shape).astype(
                     helper.tensor_dtype_to_np_dtype(getattr(TensorProto, from_type))

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

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

onnx/backend/test/case/node/ceil.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/celu.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/center_crop_pad.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/clip.py

@@ -1,6 +1,7 @@
 # Copyright (c) ONNX Project Contributors
 #
 # SPDX-License-Identifier: Apache-2.0
+from __future__ import annotations
 
 import numpy as np
 
@@ -59,6 +60,17 @@ class Clip(Base):
             name="test_clip_splitbounds",
         )
 
+        x = np.array([-2, 0, 6]).astype(np.float32)
+        y = np.array([1, 1, 1]).astype(np.float32)
+        min_val = np.float32(2)
+        max_val = np.float32(1)
+        expect(
+            node,
+            inputs=[x, min_val, max_val],
+            outputs=[y],
+            name="test_clip_min_greater_than_max",
+        )
+
     @staticmethod
     def export_clip_default() -> None:
         node = onnx.helper.make_node(

onnx/backend/test/case/node/col2im.py

@@ -1,7 +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/compress.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/concat.py

@@ -1,8 +1,9 @@
 # Copyright (c) ONNX Project Contributors
 #
 # SPDX-License-Identifier: Apache-2.0
+from __future__ import annotations
 
-from typing import Any, Dict, Sequence
+from typing import TYPE_CHECKING, Any
 
 import numpy as np
 
@@ -10,11 +11,14 @@ import onnx
 from onnx.backend.test.case.base import Base
 from onnx.backend.test.case.node import expect
 
+if TYPE_CHECKING:
+    from collections.abc import Sequence
+
 
 class Concat(Base):
     @staticmethod
     def export() -> None:
-        test_cases: Dict[str, Sequence[Any]] = {
+        test_cases: dict[str, Sequence[Any]] = {
             "1d": ([1, 2], [3, 4]),
             "2d": ([[1, 2], [3, 4]], [[5, 6], [7, 8]]),
             "3d": (

onnx/backend/test/case/node/constant.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/constantofshape.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/conv.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/convinteger.py

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