Skip to content

Commit

Permalink
quantization support in onnx (onnx#1872)
Browse files Browse the repository at this point in the history 
* add quantized ops

* shape inference for matmul ops.

* update

* add quantization parameter annotation in graph proto.

* update IR version

* update proto files

* update operator.md

* update

* fix build break.

* fix build break.

* sync and resolve comments

* revert the change by mistake.

* fix shape inference test failure.

* fix comments
  • Loading branch information
@linkerzhang
linkerzhang authored Apr 4, 2019
1 parent 030bbb8 commit 4cfa542
Show file tree
Hide file tree
Showing 11 changed files with 834 additions and 173 deletions.
18 changes: 9 additions & 9 deletions docs/Changelog.md
View file
Original file line number Diff line number Diff line change
Expand Up @@ -296,7 +296,7 @@ This version of the operator has been available since version 1 of the default O

<dl>
<dt><tt>auto_pad</tt> : string (default is NOTSET)</dt>
<dd>auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that the output size match the input.In case of odd number add the extra padding at the end for SAME_UPPER and at the beginning for SAME_LOWER. VALID mean no padding. DEPRECATION NOTE: auto_pad is only intended to support legacy uses, and for framework authors, one is explicitly encouraged to use explicit padding specified in the pads attribute.</dd>
<dd>auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that the output size match the input.In case of odd number add the extra padding at the end for SAME_UPPER and at the beginning for SAME_LOWER. VALID mean no padding.</dd>
<dt><tt>kernel_shape</tt> : list of ints (required)</dt>
<dd>The size of the kernel along each axis.</dd>
<dt><tt>pads</tt> : list of ints</dt>
Expand Down Expand Up @@ -595,7 +595,7 @@ This version of the operator has been available since version 1 of the default O

<dl>
<dt><tt>auto_pad</tt> : string (default is NOTSET)</dt>
<dd>auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that the output size match the input.In case of odd number add the extra padding at the end for SAME_UPPER and at the beginning for SAME_LOWER. VALID mean no padding. DEPRECATION NOTE: auto_pad is only intended to support legacy uses, and for framework authors, one is explicitly encouraged to use explicit padding specified in the pads attribute.</dd>
<dd>auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that the output size match the input.In case of odd number add the extra padding at the end for SAME_UPPER and at the beginning for SAME_LOWER. VALID mean no padding.</dd>
<dt><tt>dilations</tt> : list of ints</dt>
<dd>dilation value along each axis of the filter.</dd>
<dt><tt>group</tt> : int (default is 1)</dt>
Expand Down Expand Up @@ -658,7 +658,7 @@ This version of the operator has been available since version 1 of the default O

<dl>
<dt><tt>auto_pad</tt> : string (default is NOTSET)</dt>
<dd>auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that the output size match the input.In case of odd number add the extra padding at the end for SAME_UPPER and at the beginning for SAME_LOWER. VALID mean no padding. DEPRECATION NOTE: auto_pad is only intended to support legacy uses, and for framework authors, one is explicitly encouraged to use explicit padding specified in the pads attribute.</dd>
<dd>auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that the output size match the input.In case of odd number add the extra padding at the end for SAME_UPPER and at the beginning for SAME_LOWER. VALID mean no padding.</dd>
<dt><tt>dilations</tt> : list of ints</dt>
<dd>dilation value along each axis of the filter.</dd>
<dt><tt>group</tt> : int (default is 1)</dt>
Expand Down Expand Up @@ -2469,7 +2469,7 @@ This version of the operator has been available since version 1 of the default O

<dl>
<dt><tt>auto_pad</tt> : string (default is NOTSET)</dt>
<dd>auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that the output size match the input.In case of odd number add the extra padding at the end for SAME_UPPER and at the beginning for SAME_LOWER. VALID mean no padding. DEPRECATION NOTE: auto_pad is only intended to support legacy uses, and for framework authors, one is explicitly encouraged to use explicit padding specified in the pads attribute.</dd>
<dd>auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that the output size match the input.In case of odd number add the extra padding at the end for SAME_UPPER and at the beginning for SAME_LOWER. VALID mean no padding.</dd>
<dt><tt>kernel_shape</tt> : list of ints (required)</dt>
<dd>The size of the kernel along each axis.</dd>
<dt><tt>pads</tt> : list of ints</dt>
Expand Down Expand Up @@ -4785,7 +4785,7 @@ This version of the operator has been available since version 2 of the default O

<dl>
<dt><tt>auto_pad</tt> : string (default is NOTSET)</dt>
<dd>auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that the output size match the input.In case of odd number add the extra padding at the end for SAME_UPPER and at the beginning for SAME_LOWER. VALID mean no padding. DEPRECATION NOTE: auto_pad is only intended to support legacy uses, and for framework authors, one is explicitly encouraged to use explicit padding specified in the pads attribute.</dd>
<dd>auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that the output size match the input.In case of odd number add the extra padding at the end for SAME_UPPER and at the beginning for SAME_LOWER. VALID mean no padding.</dd>
<dt><tt>kernel_shape</tt> : list of ints (required)</dt>
<dd>The size of the kernel along each axis.</dd>
<dt><tt>p</tt> : int (default is 2)</dt>
Expand Down Expand Up @@ -6520,7 +6520,7 @@ This version of the operator has been available since version 7 of the default O

<dl>
<dt><tt>auto_pad</tt> : string (default is NOTSET)</dt>
<dd>auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that the output size match the input.In case of odd number add the extra padding at the end for SAME_UPPER and at the beginning for SAME_LOWER. VALID mean no padding. DEPRECATION NOTE: auto_pad is only intended to support legacy uses, and for framework authors, one is explicitly encouraged to use explicit padding specified in the pads attribute.</dd>
<dd>auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that the output size match the input.In case of odd number add the extra padding at the end for SAME_UPPER and at the beginning for SAME_LOWER. VALID mean no padding.</dd>
<dt><tt>count_include_pad</tt> : int (default is 0)</dt>
<dd>Whether include pad pixels when calculating values for the edges. Default is 0, doesn't count include pad.</dd>
<dt><tt>kernel_shape</tt> : list of ints (required)</dt>
Expand Down Expand Up @@ -7736,7 +7736,7 @@ This version of the operator has been available since version 8 of the default O

<dl>
<dt><tt>auto_pad</tt> : string (default is NOTSET)</dt>
<dd>auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that the output size match the input.In case of odd number add the extra padding at the end for SAME_UPPER and at the beginning for SAME_LOWER. VALID mean no padding. DEPRECATION NOTE: auto_pad is only intended to support legacy uses, and for framework authors, one is explicitly encouraged to use explicit padding specified in the pads attribute.</dd>
<dd>auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that the output size match the input.In case of odd number add the extra padding at the end for SAME_UPPER and at the beginning for SAME_LOWER. VALID mean no padding.</dd>
<dt><tt>kernel_shape</tt> : list of ints (required)</dt>
<dd>The size of the kernel along each axis.</dd>
<dt><tt>pads</tt> : list of ints</dt>
Expand Down Expand Up @@ -9452,7 +9452,7 @@ This version of the operator has been available since version 10 of the default

<dl>
<dt><tt>auto_pad</tt> : string (default is NOTSET)</dt>
<dd>auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that the output size match the input.In case of odd number add the extra padding at the end for SAME_UPPER and at the beginning for SAME_LOWER. VALID mean no padding. DEPRECATION NOTE: auto_pad is only intended to support legacy uses, and for framework authors, one is explicitly encouraged to use explicit padding specified in the pads attribute.</dd>
<dd>auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that the output size match the input.In case of odd number add the extra padding at the end for SAME_UPPER and at the beginning for SAME_LOWER. VALID mean no padding.</dd>
<dt><tt>ceil_mode</tt> : int (default is 0)</dt>
<dd>Wether to use ceil or floor (default) to compute the output shape.</dd>
<dt><tt>count_include_pad</tt> : int (default is 0)</dt>
Expand Down Expand Up @@ -9571,7 +9571,7 @@ This version of the operator has been available since version 10 of the default

<dl>
<dt><tt>auto_pad</tt> : string (default is NOTSET)</dt>
<dd>auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that the output size match the input.In case of odd number add the extra padding at the end for SAME_UPPER and at the beginning for SAME_LOWER. VALID mean no padding. DEPRECATION NOTE: auto_pad is only intended to support legacy uses, and for framework authors, one is explicitly encouraged to use explicit padding specified in the pads attribute.</dd>
<dd>auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that the output size match the input.In case of odd number add the extra padding at the end for SAME_UPPER and at the beginning for SAME_LOWER. VALID mean no padding.</dd>
<dt><tt>ceil_mode</tt> : int (default is 0)</dt>
<dd>Wether to use ceil or floor (default) to compute the output shape.</dd>
<dt><tt>dilations</tt> : list of ints</dt>
Expand Down
10 changes: 5 additions & 5 deletions docs/Operators.md
View file
Original file line number Diff line number Diff line change
Expand Up @@ -1036,7 +1036,7 @@ Other versions of this operator: <a href="Changelog.md#AveragePool-1">AveragePoo

<dl>
<dt><tt>auto_pad</tt> : string (default is NOTSET)</dt>
<dd>auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that the output size match the input.In case of odd number add the extra padding at the end for SAME_UPPER and at the beginning for SAME_LOWER. VALID mean no padding. DEPRECATION NOTE: auto_pad is only intended to support legacy uses, and for framework authors, one is explicitly encouraged to use explicit padding specified in the pads attribute.</dd>
<dd>auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that the output size match the input.In case of odd number add the extra padding at the end for SAME_UPPER and at the beginning for SAME_LOWER. VALID mean no padding.</dd>
<dt><tt>ceil_mode</tt> : int (default is 0)</dt>
<dd>Wether to use ceil or floor (default) to compute the output shape.</dd>
<dt><tt>count_include_pad</tt> : int (default is 0)</dt>
Expand Down Expand Up @@ -2273,7 +2273,7 @@ This version of the operator has been available since version 1 of the default O

<dl>
<dt><tt>auto_pad</tt> : string (default is NOTSET)</dt>
<dd>auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that the output size match the input.In case of odd number add the extra padding at the end for SAME_UPPER and at the beginning for SAME_LOWER. VALID mean no padding. DEPRECATION NOTE: auto_pad is only intended to support legacy uses, and for framework authors, one is explicitly encouraged to use explicit padding specified in the pads attribute.</dd>
<dd>auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that the output size match the input.In case of odd number add the extra padding at the end for SAME_UPPER and at the beginning for SAME_LOWER. VALID mean no padding.</dd>
<dt><tt>dilations</tt> : list of ints</dt>
<dd>dilation value along each axis of the filter.</dd>
<dt><tt>group</tt> : int (default is 1)</dt>
Expand Down Expand Up @@ -2456,7 +2456,7 @@ This version of the operator has been available since version 1 of the default O

<dl>
<dt><tt>auto_pad</tt> : string (default is NOTSET)</dt>
<dd>auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that the output size match the input.In case of odd number add the extra padding at the end for SAME_UPPER and at the beginning for SAME_LOWER. VALID mean no padding. DEPRECATION NOTE: auto_pad is only intended to support legacy uses, and for framework authors, one is explicitly encouraged to use explicit padding specified in the pads attribute.</dd>
<dd>auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that the output size match the input.In case of odd number add the extra padding at the end for SAME_UPPER and at the beginning for SAME_LOWER. VALID mean no padding.</dd>
<dt><tt>dilations</tt> : list of ints</dt>
<dd>dilation value along each axis of the filter.</dd>
<dt><tt>group</tt> : int (default is 1)</dt>
Expand Down Expand Up @@ -5977,7 +5977,7 @@ Other versions of this operator: <a href="Changelog.md#LpPool-1">LpPool-1</a>

<dl>
<dt><tt>auto_pad</tt> : string (default is NOTSET)</dt>
<dd>auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that the output size match the input.In case of odd number add the extra padding at the end for SAME_UPPER and at the beginning for SAME_LOWER. VALID mean no padding. DEPRECATION NOTE: auto_pad is only intended to support legacy uses, and for framework authors, one is explicitly encouraged to use explicit padding specified in the pads attribute.</dd>
<dd>auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that the output size match the input.In case of odd number add the extra padding at the end for SAME_UPPER and at the beginning for SAME_LOWER. VALID mean no padding.</dd>
<dt><tt>kernel_shape</tt> : list of ints (required)</dt>
<dd>The size of the kernel along each axis.</dd>
<dt><tt>p</tt> : int (default is 2)</dt>
Expand Down Expand Up @@ -6196,7 +6196,7 @@ Other versions of this operator: <a href="Changelog.md#MaxPool-1">MaxPool-1</a>,

<dl>
<dt><tt>auto_pad</tt> : string (default is NOTSET)</dt>
<dd>auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that the output size match the input.In case of odd number add the extra padding at the end for SAME_UPPER and at the beginning for SAME_LOWER. VALID mean no padding. DEPRECATION NOTE: auto_pad is only intended to support legacy uses, and for framework authors, one is explicitly encouraged to use explicit padding specified in the pads attribute.</dd>
<dd>auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that the output size match the input.In case of odd number add the extra padding at the end for SAME_UPPER and at the beginning for SAME_LOWER. VALID mean no padding.</dd>
<dt><tt>ceil_mode</tt> : int (default is 0)</dt>
<dd>Wether to use ceil or floor (default) to compute the output shape.</dd>
<dt><tt>dilations</tt> : list of ints</dt>
Expand Down
Loading

0 comments on commit 4cfa542

Please sign in to comment.