Skip to content

Commit da7c50c

Browse files
authored
ONNX does not maintain versions for experimental ops (#1696)
* fix the versioning info for exp ops, which do not have version maintained in onnx. * remove the version 9 for dynamicslice, which is misleading partners as exp ops do not versioning maintained actually. * fix build error * update the changelog.md
1 parent 0c8d857 commit da7c50c

File tree

5 files changed

+110
-127
lines changed

5 files changed

+110
-127
lines changed

docs/Changelog.md

Lines changed: 78 additions & 91 deletions
Original file line numberDiff line numberDiff line change
@@ -13,8 +13,7 @@
1313

1414
#### Version
1515

16-
This version of the operator has been available since version 1 of the default ONNX operator set.
17-
16+
No versioning maintained for experimental ops.
1817
#### Inputs (1 - ∞)
1918

2019
<dl>
@@ -143,8 +142,7 @@ This version of the operator has been available since version 1 of the default O
143142

144143
#### Version
145144

146-
This version of the operator has been available since version 1 of the default ONNX operator set.
147-
145+
No versioning maintained for experimental ops.
148146
#### Attributes
149147

150148
<dl>
@@ -646,8 +644,7 @@ This version of the operator has been available since version 1 of the default O
646644

647645
#### Version
648646

649-
This version of the operator has been available since version 1 of the default ONNX operator set.
650-
647+
No versioning maintained for experimental ops.
651648
#### Attributes
652649

653650
<dl>
@@ -812,8 +809,7 @@ This version of the operator has been available since version 1 of the default O
812809

813810
#### Version
814811

815-
This version of the operator has been available since version 1 of the default ONNX operator set.
816-
812+
No versioning maintained for experimental ops.
817813
#### Attributes
818814

819815
<dl>
@@ -990,6 +986,72 @@ This version of the operator has been available since version 1 of the default O
990986
<dd>Constrain input and output types to float tensors.</dd>
991987
</dl>
992988

989+
### <a name="DynamicSlice-1"></a>**DynamicSlice-1**</a>
990+
991+
Produces a slice of the input tensor along multiple axes. Similar to numpy:
992+
https://docs.scipy.org/doc/numpy/reference/arrays.indexing.html
993+
Slices uses `axes`, `starts` and `ends` inputs to specify the start and end
994+
dimension for each axis in the list of axes, it uses this information to
995+
slice the input `data` tensor. If a negative value is passed for any of the
996+
start or end indices, it represent number of elements before the end of that
997+
dimension. If the value passed to start or end is larger than the `n` (the
998+
number of elements in this dimension), it represents `n`. For slicing to the
999+
end of a dimension with unknown size, it is recommended to pass in `INT_MAX`.
1000+
If `axes` are omitted, they are set to `[0, ..., ndim-1]`.
1001+
Example 1:
1002+
data = [
1003+
[1, 2, 3, 4],
1004+
[5, 6, 7, 8],
1005+
]
1006+
axes = [0, 1]
1007+
starts = [1, 0]
1008+
ends = [2, 3]
1009+
result = [
1010+
[5, 6, 7],
1011+
]
1012+
Example 2:
1013+
data = [
1014+
[1, 2, 3, 4],
1015+
[5, 6, 7, 8],
1016+
]
1017+
starts = [0, 1]
1018+
ends = [-1, 1000]
1019+
result = [
1020+
[2, 3, 4],
1021+
]
1022+
1023+
#### Version
1024+
1025+
No versioning maintained for experimental ops.
1026+
#### Inputs (3 - 4)
1027+
1028+
<dl>
1029+
<dt><tt>data</tt> : T</dt>
1030+
<dd>Tensor of data to extract slices from.</dd>
1031+
<dt><tt>starts</tt> : Tind</dt>
1032+
<dd>1-D tensor of starting indices of corresponding axis in `axes`</dd>
1033+
<dt><tt>ends</tt> : Tind</dt>
1034+
<dd>1-D tensor of ending indices (exclusive) of corresponding axis in axes</dd>
1035+
<dt><tt>axes</tt> (optional) : Tind</dt>
1036+
<dd>1-D tensor of axes that `starts` and `ends` apply to.</dd>
1037+
</dl>
1038+
1039+
#### Outputs
1040+
1041+
<dl>
1042+
<dt><tt>output</tt> : T</dt>
1043+
<dd>Sliced data tensor.</dd>
1044+
</dl>
1045+
1046+
#### Type Constraints
1047+
1048+
<dl>
1049+
<dt><tt>T</tt> : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)</dt>
1050+
<dd>Constrain input and output types to all tensor types.</dd>
1051+
<dt><tt>Tind</tt> : tensor(int32), tensor(int64)</dt>
1052+
<dd>Constrain indices to integer types</dd>
1053+
</dl>
1054+
9931055
### <a name="Elu-1"></a>**Elu-1**</a>
9941056

9951057
Elu takes one input data (Tensor<T>) and produces one output data
@@ -1334,8 +1396,7 @@ This version of the operator has been available since version 1 of the default O
13341396

13351397
#### Version
13361398

1337-
This version of the operator has been available since version 1 of the default ONNX operator set.
1338-
1399+
No versioning maintained for experimental ops.
13391400
#### Attributes
13401401

13411402
<dl>
@@ -1508,8 +1569,7 @@ This version of the operator has been available since version 1 of the default O
15081569

15091570
#### Version
15101571

1511-
This version of the operator has been available since version 1 of the default ONNX operator set.
1512-
1572+
No versioning maintained for experimental ops.
15131573
#### Attributes
15141574

15151575
<dl>
@@ -1859,8 +1919,7 @@ This version of the operator has been available since version 1 of the default O
18591919

18601920
#### Version
18611921

1862-
This version of the operator has been available since version 1 of the default ONNX operator set.
1863-
1922+
No versioning maintained for experimental ops.
18641923
#### Attributes
18651924

18661925
<dl>
@@ -3082,8 +3141,7 @@ This version of the operator has been available since version 1 of the default O
30823141

30833142
#### Version
30843143

3085-
This version of the operator has been available since version 1 of the default ONNX operator set.
3086-
3144+
No versioning maintained for experimental ops.
30873145
#### Attributes
30883146

30893147
<dl>
@@ -4043,8 +4101,7 @@ This version of the operator has been available since version 1 of the default O
40434101

40444102
#### Version
40454103

4046-
This version of the operator has been available since version 1 of the default ONNX operator set.
4047-
4104+
No versioning maintained for experimental ops.
40484105
#### Attributes
40494106

40504107
<dl>
@@ -4081,8 +4138,7 @@ This version of the operator has been available since version 1 of the default O
40814138

40824139
#### Version
40834140

4084-
This version of the operator has been available since version 1 of the default ONNX operator set.
4085-
4141+
No versioning maintained for experimental ops.
40864142
#### Attributes
40874143

40884144
<dl>
@@ -4736,8 +4792,7 @@ This version of the operator has been available since version 1 of the default O
47364792

47374793
#### Version
47384794

4739-
This version of the operator has been available since version 1 of the default ONNX operator set.
4740-
4795+
No versioning maintained for experimental ops.
47414796
#### Attributes
47424797

47434798
<dl>
@@ -4955,8 +5010,7 @@ This version of the operator has been available since version 1 of the default O
49555010

49565011
#### Version
49575012

4958-
This version of the operator has been available since version 1 of the default ONNX operator set.
4959-
5013+
No versioning maintained for experimental ops.
49605014
#### Attributes
49615015

49625016
<dl>
@@ -8581,73 +8635,6 @@ This version of the operator has been available since version 9 of the default O
85818635
<dd>Constrain input and output types to float tensors.</dd>
85828636
</dl>
85838637

8584-
### <a name="DynamicSlice-9"></a>**DynamicSlice-9**</a>
8585-
8586-
Produces a slice of the input tensor along multiple axes. Similar to numpy:
8587-
https://docs.scipy.org/doc/numpy/reference/arrays.indexing.html
8588-
Slices uses `axes`, `starts` and `ends` inputs to specify the start and end
8589-
dimension for each axis in the list of axes, it uses this information to
8590-
slice the input `data` tensor. If a negative value is passed for any of the
8591-
start or end indices, it represent number of elements before the end of that
8592-
dimension. If the value passed to start or end is larger than the `n` (the
8593-
number of elements in this dimension), it represents `n`. For slicing to the
8594-
end of a dimension with unknown size, it is recommended to pass in `INT_MAX`.
8595-
If `axes` are omitted, they are set to `[0, ..., ndim-1]`.
8596-
Example 1:
8597-
data = [
8598-
[1, 2, 3, 4],
8599-
[5, 6, 7, 8],
8600-
]
8601-
axes = [0, 1]
8602-
starts = [1, 0]
8603-
ends = [2, 3]
8604-
result = [
8605-
[5, 6, 7],
8606-
]
8607-
Example 2:
8608-
data = [
8609-
[1, 2, 3, 4],
8610-
[5, 6, 7, 8],
8611-
]
8612-
starts = [0, 1]
8613-
ends = [-1, 1000]
8614-
result = [
8615-
[2, 3, 4],
8616-
]
8617-
8618-
#### Version
8619-
8620-
This version of the operator has been available since version 9 of the default ONNX operator set.
8621-
8622-
#### Inputs (3 - 4)
8623-
8624-
<dl>
8625-
<dt><tt>data</tt> : T</dt>
8626-
<dd>Tensor of data to extract slices from.</dd>
8627-
<dt><tt>starts</tt> : Tind</dt>
8628-
<dd>1-D tensor of starting indices of corresponding axis in `axes`</dd>
8629-
<dt><tt>ends</tt> : Tind</dt>
8630-
<dd>1-D tensor of ending indices (exclusive) of corresponding axis in axes</dd>
8631-
<dt><tt>axes</tt> (optional) : Tind</dt>
8632-
<dd>1-D tensor of axes that `starts` and `ends` apply to.</dd>
8633-
</dl>
8634-
8635-
#### Outputs
8636-
8637-
<dl>
8638-
<dt><tt>output</tt> : T</dt>
8639-
<dd>Sliced data tensor.</dd>
8640-
</dl>
8641-
8642-
#### Type Constraints
8643-
8644-
<dl>
8645-
<dt><tt>T</tt> : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128)</dt>
8646-
<dd>Constrain input and output types to all tensor types.</dd>
8647-
<dt><tt>Tind</tt> : tensor(int32), tensor(int64)</dt>
8648-
<dd>Constrain indices to integer types</dd>
8649-
</dl>
8650-
86518638
### <a name="Erf-9"></a>**Erf-9**</a>
86528639

86538640
Computes the error function of the given input tensor element-wise.

docs/Operators.md

Lines changed: 12 additions & 24 deletions
Original file line numberDiff line numberDiff line change
@@ -12008,8 +12008,7 @@ expect(node, inputs=[x, y], outputs=[z],
1200812008

1200912009
#### Version
1201012010

12011-
This version of the operator has been available since version 1 of the default ONNX operator set.
12012-
12011+
No versioning maintained for experimental ops.
1201312012
#### Inputs (1 - &#8734;)
1201412013

1201512014
<dl>
@@ -12040,8 +12039,7 @@ This version of the operator has been available since version 1 of the default O
1204012039

1204112040
#### Version
1204212041

12043-
This version of the operator has been available since version 1 of the default ONNX operator set.
12044-
12042+
No versioning maintained for experimental ops.
1204512043
#### Attributes
1204612044

1204712045
<dl>
@@ -12096,8 +12094,7 @@ This version of the operator has been available since version 1 of the default O
1209612094

1209712095
#### Version
1209812096

12099-
This version of the operator has been available since version 1 of the default ONNX operator set.
12100-
12097+
No versioning maintained for experimental ops.
1210112098
#### Attributes
1210212099

1210312100
<dl>
@@ -12145,8 +12142,7 @@ This version of the operator has been available since version 1 of the default O
1214512142

1214612143
#### Version
1214712144

12148-
This version of the operator has been available since version 1 of the default ONNX operator set.
12149-
12145+
No versioning maintained for experimental ops.
1215012146
#### Attributes
1215112147

1215212148
<dl>
@@ -12214,8 +12210,7 @@ This version of the operator has been available since version 1 of the default O
1221412210

1221512211
#### Version
1221612212

12217-
This version of the operator has been available since version 9 of the default ONNX operator set.
12218-
12213+
No versioning maintained for experimental ops.
1221912214
#### Inputs (3 - 4)
1222012215

1222112216
<dl>
@@ -12373,8 +12368,7 @@ expect(node, inputs=[x, starts, ends, axes], outputs=[y],
1237312368

1237412369
#### Version
1237512370

12376-
This version of the operator has been available since version 1 of the default ONNX operator set.
12377-
12371+
No versioning maintained for experimental ops.
1237812372
#### Attributes
1237912373

1238012374
<dl>
@@ -12414,8 +12408,7 @@ This version of the operator has been available since version 1 of the default O
1241412408

1241512409
#### Version
1241612410

12417-
This version of the operator has been available since version 1 of the default ONNX operator set.
12418-
12411+
No versioning maintained for experimental ops.
1241912412
#### Attributes
1242012413

1242112414
<dl>
@@ -12458,8 +12451,7 @@ This version of the operator has been available since version 1 of the default O
1245812451

1245912452
#### Version
1246012453

12461-
This version of the operator has been available since version 1 of the default ONNX operator set.
12462-
12454+
No versioning maintained for experimental ops.
1246312455
#### Attributes
1246412456

1246512457
<dl>
@@ -12499,8 +12491,7 @@ This version of the operator has been available since version 1 of the default O
1249912491

1250012492
#### Version
1250112493

12502-
This version of the operator has been available since version 1 of the default ONNX operator set.
12503-
12494+
No versioning maintained for experimental ops.
1250412495
#### Attributes
1250512496

1250612497
<dl>
@@ -12539,8 +12530,7 @@ This version of the operator has been available since version 1 of the default O
1253912530

1254012531
#### Version
1254112532

12542-
This version of the operator has been available since version 1 of the default ONNX operator set.
12543-
12533+
No versioning maintained for experimental ops.
1254412534
#### Attributes
1254512535

1254612536
<dl>
@@ -12578,8 +12568,7 @@ This version of the operator has been available since version 1 of the default O
1257812568

1257912569
#### Version
1258012570

12581-
This version of the operator has been available since version 1 of the default ONNX operator set.
12582-
12571+
No versioning maintained for experimental ops.
1258312572
#### Attributes
1258412573

1258512574
<dl>
@@ -12619,8 +12608,7 @@ This version of the operator has been available since version 1 of the default O
1261912608

1262012609
#### Version
1262112610

12622-
This version of the operator has been available since version 1 of the default ONNX operator set.
12623-
12611+
No versioning maintained for experimental ops.
1262412612
#### Attributes
1262512613

1262612614
<dl>

0 commit comments

Comments
 (0)