onnx 1.16.2__cp38-cp38-win32.whl → 1.17.0__cp38-cp38-win32.whl

onnx/reference/ops/op_pool_common.py

@@ -1,10 +1,11 @@
 # Copyright (c) ONNX Project Contributors
 #
 # SPDX-License-Identifier: Apache-2.0
+from __future__ import annotations
 
 import itertools
 import math
-from typing import Sequence, Tuple, Union
+from typing import Sequence
 
 import numpy as np
 
@@ -54,9 +55,9 @@ def get_output_shape_explicit_padding(
     input_spatial_shape: Sequence[int],
     kernel_spatial_shape: Sequence[int],
     strides_spatial: Sequence[int],
-    dilations: Union[Sequence[int], None] = None,
+    dilations: Sequence[int] | None = None,
     ceil_mode: bool = False,
-) -> Tuple[Sequence[int], Sequence[int]]:
+) -> tuple[Sequence[int], Sequence[int]]:
     """Compute output shape according to:
     https://pytorch.org/docs/stable/generated/torch.nn.MaxPool1d.html?highlight=max+pool#torch.nn.MaxPool1d
     Pads are used to calculate output shape. Use output shape in turn to calculate the actual pads
@@ -165,8 +166,8 @@ def pool(
     strides: Sequence[int],
     out_shape: Sequence[int],
     pooling_type: str,
-    pads: Union[Sequence[int], None] = None,
-    dilations: Union[Sequence[int], None] = None,
+    pads: Sequence[int] | None = None,
+    dilations: Sequence[int] | None = None,
     count_include_pad: int = 0,
     p: int = 1,
 ) -> np.ndarray:
@@ -248,7 +249,7 @@ def pool(
             y[shape] = f(window_vals)
         else:
             y[shape] = f(window_vals[np.where(~np.isnan(window_vals))])
-    return y
+    return y.astype(padded.dtype)
 
 
 class CommonPool(OpRun):

onnx/reference/ops/op_pow.py

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

onnx/reference/ops/op_prelu.py

@@ -1,7 +1,7 @@
 # Copyright (c) ONNX Project Contributors
 
 # SPDX-License-Identifier: Apache-2.0
-
+from __future__ import annotations
 
 import numpy as np
 
@@ -12,7 +12,7 @@ class PRelu(OpRun):
     def _run(self, x, slope):  # type: ignore
         try:
             return (np.where(x > 0, x, x * slope).astype(x.dtype),)
-        except ValueError as e:
+        except ValueError:
             # Broadcast did not work according to numpy.
             # The logic is then the following, if slope has d elements,
             # the following code is looking for d in x.shape. If it is found
@@ -31,4 +31,4 @@ class PRelu(OpRun):
                 if n == 1:
                     xs = x * slope.reshape(tuple(new_shape))
                     return (np.where(x > 0, x, xs).astype(x.dtype),)
-            raise e
+            raise

onnx/reference/ops/op_qlinear_conv.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/reference/ops/op_qlinear_matmul.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/reference/ops/op_quantize_linear.py

@@ -10,13 +10,7 @@ from typing import ClassVar
 import numpy as np
 
 from onnx import TensorProto, subbyte
-from onnx.helper import (
-    float32_to_float8e4m3,
-    float32_to_float8e5m2,
-    np_dtype_to_tensor_dtype,
-    tensor_dtype_to_np_dtype,
-)
-from onnx.reference.custom_element_types import (
+from onnx._custom_element_types import (
     float8e4m3fn,
     float8e4m3fnuz,
     float8e5m2,
@@ -24,6 +18,12 @@ from onnx.reference.custom_element_types import (
     int4,
     uint4,
 )
+from onnx.helper import (
+    float32_to_float8e4m3,
+    float32_to_float8e5m2,
+    np_dtype_to_tensor_dtype,
+    tensor_dtype_to_np_dtype,
+)
 from onnx.reference.op_run import OpRun
 
 
@@ -68,7 +68,7 @@ def reshape_input(
     if (
         shape[axis] != value.shape[axis]
     ):  # block_size does not divide x, handle the remainder block
-        value = value.take(indices=range(0, shape[axis]), axis=axis)
+        value = value.take(indices=range(shape[axis]), axis=axis)
     if value.shape != shape:
         raise ValueError(
             "Invalid shapes for Blocked Quantization. Input 2 shape should identical to Input 1 shape, except for one dimension, in which blocking is performed"
@@ -221,4 +221,10 @@ class QuantizeLinear_19(_CommonQuantizeLinear):
 class QuantizeLinear_21(_CommonQuantizeLinear):
     def _run(self, *args, axis=None, saturate=None, block_size=None, output_dtype=None):  # type: ignore
         # args: x, y_scale, zero_point
-        return super()._run(*args, axis=axis, saturate=saturate, block_size=block_size, output_dtype=output_dtype)  # type: ignore
+        return super()._run(
+            *args,
+            axis=axis,
+            saturate=saturate,
+            block_size=block_size,
+            output_dtype=output_dtype,
+        )  # type: ignore

onnx/reference/ops/op_random_normal.py

@@ -1,7 +1,7 @@
 # Copyright (c) ONNX Project Contributors
 
 # SPDX-License-Identifier: Apache-2.0
-
+from __future__ import annotations
 
 from onnx.reference.ops._op_common_random import _CommonRandom
 

onnx/reference/ops/op_random_normal_like.py

@@ -1,7 +1,7 @@
 # Copyright (c) ONNX Project Contributors
 
 # SPDX-License-Identifier: Apache-2.0
-
+from __future__ import annotations
 
 from onnx.helper import np_dtype_to_tensor_dtype
 from onnx.reference.ops._op_common_random import _CommonRandom

onnx/reference/ops/op_random_uniform.py

@@ -1,7 +1,7 @@
 # Copyright (c) ONNX Project Contributors
 
 # SPDX-License-Identifier: Apache-2.0
-
+from __future__ import annotations
 
 from onnx.reference.ops._op_common_random import _CommonRandom
 

onnx/reference/ops/op_random_uniform_like.py

@@ -1,7 +1,7 @@
 # Copyright (c) ONNX Project Contributors
 
 # SPDX-License-Identifier: Apache-2.0
-
+from __future__ import annotations
 
 from onnx.helper import np_dtype_to_tensor_dtype
 from onnx.reference.ops._op_common_random import _CommonRandom

onnx/reference/ops/op_range.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/reference/ops/op_reciprocal.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/reference/ops/op_reduce_l1.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/reference/ops/op_reduce_l2.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/reference/ops/op_reduce_log_sum.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/reference/ops/op_reduce_log_sum_exp.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/reference/ops/op_reduce_max.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/reference/ops/op_reduce_mean.py

@@ -1,7 +1,7 @@
 # Copyright (c) ONNX Project Contributors
 
 # SPDX-License-Identifier: Apache-2.0
-
+from __future__ import annotations
 
 import numpy as np
 
@@ -30,8 +30,8 @@ class ReduceMean_18(OpRunReduceNumpy):
             if keepdims == 0 and not isinstance(res, np.ndarray):
                 # The runtime must return a numpy array of a single float.
                 res = np.array(res)
-            return (res,)  # type: ignore
         except TypeError as e:
             raise TypeError(
                 f"Unable to reduce shape {data.shape!r} with axes={axes!r} and keepdims={keepdims}."
             ) from e
+        return (res,)  # type: ignore

onnx/reference/ops/op_reduce_min.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/reference/ops/op_reduce_prod.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/reference/ops/op_reduce_sum.py

@@ -1,7 +1,7 @@
 # Copyright (c) ONNX Project Contributors
 
 # SPDX-License-Identifier: Apache-2.0
-
+from __future__ import annotations
 
 import numpy as np
 
@@ -28,7 +28,7 @@ class ReduceSum_13(OpRunReduceNumpy):
             if keepdims == 0 and not isinstance(res, np.ndarray):
                 # The runtime must return a numpy array of a single float.
                 res = np.array(res)
-            return (res,)  # type: ignore
+            return (res,)  # type: ignore  # noqa: TRY300
         except TypeError as e:
             raise TypeError(
                 f"Unable to reduce shape {x.shape!r} with axes={axes!r} and keepdims={keepdims}."

onnx/reference/ops/op_reduce_sum_square.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/reference/ops/op_regex_full_match.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/reference/ops/op_relu.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/reference/ops/op_reshape.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/reference/ops/op_reverse_sequence.py

@@ -1,7 +1,7 @@
 # Copyright (c) ONNX Project Contributors
 
 # SPDX-License-Identifier: Apache-2.0
-
+from __future__ import annotations
 
 from onnx.reference.op_run import OpRun
 

onnx/reference/ops/op_rnn.py

@@ -1,7 +1,7 @@
 # Copyright (c) ONNX Project Contributors
 
 # SPDX-License-Identifier: Apache-2.0
-
+from __future__ import annotations
 
 import numpy as np
 
@@ -52,12 +52,14 @@ class CommonRNN(OpRun):
                 self.activations[1],  # type: ignore
                 (
                     self.activation_alpha[1]  # type: ignore
-                    if self.activation_alpha is not None and len(self.activation_alpha) > 1  # type: ignore
+                    if self.activation_alpha is not None
+                    and len(self.activation_alpha) > 1  # type: ignore
                     else None
                 ),
                 (
                     self.activation_beta[1]  # type: ignore
-                    if self.activation_beta is not None and len(self.activation_beta) > 1  # type: ignore
+                    if self.activation_beta is not None
+                    and len(self.activation_beta) > 1  # type: ignore
                     else None
                 ),
             )
@@ -97,11 +99,11 @@ class CommonRNN(OpRun):
         B=None,
         sequence_lens=None,
         initial_h=None,
-        activation_alpha=None,
-        activation_beta=None,
-        activations=None,
-        clip=None,
-        direction=None,
+        activation_alpha=None,  # noqa: ARG002
+        activation_beta=None,  # noqa: ARG002
+        activations=None,  # noqa: ARG002
+        clip=None,  # noqa: ARG002
+        direction=None,  # noqa: ARG002
         hidden_size=None,
         layout=None,
     ):

onnx/reference/ops/op_roi_align.py

@@ -1,9 +1,7 @@
 # Copyright (c) ONNX Project Contributors
 
 # SPDX-License-Identifier: Apache-2.0
-
-
-from typing import Tuple
+from __future__ import annotations
 
 import numpy as np
 
@@ -117,7 +115,7 @@ class RoiAlign(OpRun):
 
     @staticmethod
     def roi_align_forward(  # type: ignore
-        output_shape: Tuple[int, int, int, int],
+        output_shape: tuple[int, int, int, int],
         bottom_data,
         spatial_scale,
         height: int,
@@ -265,7 +263,9 @@ class RoiAlign(OpRun):
         sampling_ratio=None,
         spatial_scale=None,
     ):
-        coordinate_transformation_mode = coordinate_transformation_mode or self.coordinate_transformation_mode  # type: ignore
+        coordinate_transformation_mode = (
+            coordinate_transformation_mode or self.coordinate_transformation_mode
+        )  # type: ignore
         mode = mode or self.mode  # type: ignore
         output_height = output_height or self.output_height  # type: ignore
         output_width = output_width or self.output_width  # type: ignore

onnx/reference/ops/op_round.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/reference/ops/op_scan.py

@@ -1,7 +1,7 @@
 # Copyright (c) ONNX Project Contributors
 
 # SPDX-License-Identifier: Apache-2.0
-
+from __future__ import annotations
 
 import numpy as np
 
@@ -99,13 +99,13 @@ class Scan(OpRun):
     def _run(  # type:ignore
         self,
         *args,
-        body=None,
-        num_scan_inputs=None,
-        scan_input_axes=None,
-        scan_input_directions=None,
-        scan_output_axes=None,
-        scan_output_directions=None,
-        attributes=None,
+        body=None,  # noqa: ARG002
+        num_scan_inputs=None,  # noqa: ARG002
+        scan_input_axes=None,  # noqa: ARG002
+        scan_input_directions=None,  # noqa: ARG002
+        scan_output_axes=None,  # noqa: ARG002
+        scan_output_directions=None,  # noqa: ARG002
+        attributes=None,  # noqa: ARG002
     ):
         # TODO: support overridden attributes.
         (

onnx/reference/ops/op_scatter_elements.py

@@ -1,7 +1,7 @@
 # Copyright (c) ONNX Project Contributors
 
 # SPDX-License-Identifier: Apache-2.0
-
+from __future__ import annotations
 
 import numpy as np
 
@@ -35,7 +35,7 @@ def scatter_elements(data, indices, updates, axis=0, reduction=None):  # type: i
 
     else:
 
-        def f(x, y):
+        def f(x, y):  # noqa: ARG001
             return y
 
     if axis < 0:
@@ -88,55 +88,24 @@ def scatter_elements(data, indices, updates, axis=0, reduction=None):  # type: i
                         )
         return scattered
 
-    idx_xsection_shape = indices.shape[:axis] + indices.shape[axis + 1 :]
-
-    def make_slice(arr, axis, i):  # type: ignore
-        slc = [slice(None)] * arr.ndim
-        slc[axis] = i
-        return slc
-
-    def unpack(packed):  # type: ignore
-        unpacked = packed[0]
-        for i in range(1, len(packed)):
-            unpacked = unpacked, packed[i]
-        return unpacked
-
-    # We use indices and axis parameters to create idx
-    # idx is in a form that can be used as a NumPy advanced
-    # indices for scattering of updates param. in data
-    idx = [
-        [
-            unpack(np.indices(idx_xsection_shape).reshape(indices.ndim - 1, -1)),
-            indices[tuple(make_slice(indices, axis, i))].reshape(1, -1)[0],
-        ]
-        for i in range(indices.shape[axis])
-    ]
-    idx = list(np.concatenate(idx, axis=1))
-    idx.insert(axis, idx.pop())
-
-    # updates_idx is a NumPy advanced indices for indexing
-    # of elements in the updates
-    updates_idx = list(idx)
-    updates_idx.pop(axis)
-    updates_idx.insert(  # type: ignore
-        axis,
-        np.repeat(np.arange(indices.shape[axis]), np.prod(idx_xsection_shape)),  # type: ignore
-    )
+    if len(indices.shape) == 4:
+        scattered = np.copy(data)
+        for a in range(indices.shape[0]):
+            for i in range(indices.shape[1]):
+                for j in range(indices.shape[2]):
+                    for k in range(indices.shape[3]):
+                        index = [a, i, j, k]
+                        index[axis] = indices[a, i, j, k]
+                        tuple_index = tuple(index)
+                        scattered[tuple_index] = f(
+                            scattered[tuple_index],
+                            updates[a, i, j, k],
+                        )
+        return scattered
 
-    scattered = np.copy(data)
-    if reduction == "min":
-        scattered[tuple(idx)] = np.minimum(
-            scattered[tuple(idx)], updates[tuple(updates_idx)]
-        )
-    elif reduction == "max":
-        scattered[tuple(idx)] = np.maximum(
-            scattered[tuple(idx)], updates[tuple(updates_idx)]
-        )
-    elif reduction == "add":
-        scattered[tuple(idx)] += updates[tuple(updates_idx)]
-    else:
-        scattered[tuple(idx)] = updates[tuple(updates_idx)]
-    return scattered
+    raise NotImplementedError(
+        f"ScatterND is not implement for indices.shape={indices.shape} and axis={axis}."
+    )
 
 
 class ScatterElements(OpRun):

onnx/reference/ops/op_scatternd.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/reference/ops/op_selu.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/reference/ops/op_sequence_at.py

@@ -1,7 +1,7 @@
 # Copyright (c) ONNX Project Contributors
 
 # SPDX-License-Identifier: Apache-2.0
-
+from __future__ import annotations
 
 from onnx.reference.op_run import OpRun
 

onnx/reference/ops/op_sequence_construct.py

@@ -1,7 +1,7 @@
 # Copyright (c) ONNX Project Contributors
 
 # SPDX-License-Identifier: Apache-2.0
-
+from __future__ import annotations
 
 from onnx.reference.op_run import OpRun
 

onnx/reference/ops/op_sequence_empty.py

@@ -1,11 +1,11 @@
 # Copyright (c) ONNX Project Contributors
 
 # SPDX-License-Identifier: Apache-2.0
-
+from __future__ import annotations
 
 from onnx.reference.op_run import OpRun
 
 
 class SequenceEmpty(OpRun):
-    def _run(self, dtype=None):  # type: ignore
+    def _run(self, dtype=None):  # type: ignore  # noqa: ARG002
         return ([],)

onnx/reference/ops/op_sequence_erase.py

@@ -1,7 +1,7 @@
 # Copyright (c) ONNX Project Contributors
 
 # SPDX-License-Identifier: Apache-2.0
-
+from __future__ import annotations
 
 from onnx.reference.op_run import OpRun
 

onnx/reference/ops/op_sequence_insert.py

@@ -1,9 +1,9 @@
 # Copyright (c) ONNX Project Contributors
 
 # SPDX-License-Identifier: Apache-2.0
+from __future__ import annotations
 
-
-from typing import Any, List, Optional, Union
+from typing import Any
 
 import numpy as np
 
@@ -11,12 +11,12 @@ from onnx.reference.op_run import OpRun
 
 
 def sequence_insert_reference_implementation(
-    sequence: Union[List[Any], np.ndarray],
+    sequence: list[Any] | np.ndarray,
     tensor: np.ndarray,
-    position: Optional[np.ndarray] = None,
-) -> List[Any]:
+    position: np.ndarray | None = None,
+) -> list[Any]:
     # make a copy of input sequence
-    seq: List[Any] = []
+    seq: list[Any] = []
     if sequence is not None and (
         not isinstance(sequence, np.ndarray) or len(sequence.shape) > 0
     ):