add onnx export module, test=develop #27831

Channingss · 2020-10-12T02:45:15Z

PR types

New features

PR changes

APIs

Describe

Add onnx export module by lazy import paddle2onnx.

API中文文档预览：

暂缺API英文文档预览

paddle-bot-old · 2020-10-12T02:45:20Z

Thanks for your contribution!
Please wait for the result of CI firstly. See Paddle CI Manual for details.

ZeyuChen

文档方面需要使用最新的API，去掉fluid

ZeyuChen · 2020-10-13T12:52:27Z

python/paddle/onnx/export.py

+                    return x * y, x / y
+
+        def export_with_input_spec():
+            paddle.fluid.enable_dygraph() 


没有使用最新版本的API

ZeyuChen · 2020-10-13T12:53:17Z

python/paddle/onnx/export.py

+        def export_with_input_variable():
+            paddle.fluid.enable_dygraph() 
+            model = Model()
+            x =  paddle.fluid.dygraph.to_variable(np.array([1]).astype('float32'), name='x')


后面没有to_variable方法，统一使用paddle.to_tensor

ZeyuChen · 2020-10-13T12:57:55Z

python/paddle/onnx/export.py

+__all__ = ['export']
+
+
+def export(layer, save_file, input_spec=None, opset_version=9, **kwargs):


竞品的函数是否有返回值？还是只是到处文件就结束？

看了一下torch，mxnet和oneflow都是写到文件里面，没有返回onnx的proto对象。

… onnx

ZeyuChen

整体修改注释和细节问题。

ZeyuChen · 2020-11-11T09:23:35Z

python/paddle/fluid/tests/unittests/test_onnx_export.py

+
+    def test_prune_graph(self):
+        model = Logic()
+        # prune model with input_spec and output_spec, which need to static and run model before export.


代码注释请使用标注的英文语法注释，首字母大写

ZeyuChen · 2020-11-11T09:28:38Z

python/paddle/onnx/export.py

+        raise ValueError(
+            "The input path MUST be format of dirname/file_prefix "
+            "[dirname\\file_prefix in Windows system], but received "
+            "file_prefix is empty string.")


这里你的warning对path做了重命名为file_prefix，其实用户不知道这个变量是啥，用户只看到path这个变量。

所以考虑修改下这个ERROR让他更有提示意义

ZeyuChen · 2020-11-11T09:30:09Z

python/paddle/onnx/export.py

+    """
+    Export Layer as ONNX format model, which can be used for inference.
+    Now, it supports a limited operater set and dynamic models.(e.g., MobileNet.)
+    More features and introduction, Please reference `Github <https://github.com/PaddlePaddle/paddle2onnx>`_ .


注释这里
Please reference xxxx -> Please refer to xxxx

ZeyuChen · 2020-11-11T09:30:51Z

python/paddle/onnx/export.py

+
+def export(layer, path, input_spec=None, opset_version=9, **configs):
+    """
+    Export Layer as ONNX format model, which can be used for inference.


Export Layer as -> Export Layer to

which can be used for inference with onnxruntime.

ZeyuChen · 2020-11-11T09:32:40Z

python/paddle/onnx/export.py

+def export(layer, path, input_spec=None, opset_version=9, **configs):
+    """
+    Export Layer as ONNX format model, which can be used for inference.
+    Now, it supports a limited operater set and dynamic models.(e.g., MobileNet.)


去掉这行注释，改为参考某一个paddle2onnx某个固定的README，了解最新支持的model zoo

直接删除了这一段描述，准备把op和模型支持的列表到时候写到教程里面。另外现在model_zoo还没有动态图的doc，等完善了再加上

ZeyuChen · 2020-11-11T09:36:17Z

python/paddle/onnx/export.py

+            method, which can be described by InputSpec or example Tensor. If None, all input variables of 
+            the original Layer's forward method would be the inputs of the exported ``ONNX`` model. Default None.
+        opset_version(int, optional): Opset version of exported ONNX model.
+            Now, stable supported opset version include 9, 10, 11. Default 9.


Default 9 -> Default: 9

ZeyuChen · 2020-11-11T09:40:37Z

python/paddle/onnx/export.py

+                def forward(self, x):
+                    return self._linear(x)
+
+            # export model with InputSpec, which supports set dynamic shape for inputs.


注释中的注释也同意下语法规范。对于Paddle内部变量，需要用``阔起来，如InputSpec
简化下语法表达。

Export model with InputSpec to support dynamic input shape.

尽量用短句注释

ZeyuChen · 2020-11-11T09:45:37Z

python/paddle/onnx/export.py

+                    else:
+                        return y
+
+            # export model with Tensor, which supports prune model by set 'output_spec' with output of model.


Export model with paddle.Tensor as input_spec, and model output as output_spec, onnx exporter will automatically prune out the inference model.

ZeyuChen · 2020-11-11T09:46:18Z

python/paddle/__init__.py

@@ -271,6 +271,8 @@
 from . import static
 from . import amp



为什么这里要莫名的空行？

去掉了空行

ZeyuChen · 2020-11-11T09:46:27Z

python/paddle/__init__.py

@@ -271,6 +271,8 @@
 from . import static
 from . import amp



为什么这里要莫名的空行

ZeyuChen

只剩一点英文小意见。

ZeyuChen · 2020-11-11T12:10:50Z

python/paddle/onnx/export.py

+def export(layer, path, input_spec=None, opset_version=9, **configs):
+    """
+    Export Layer to ONNX format, which can be inferenced by onnxruntime or other backends.
+    More details, Please refer to `paddle2onnx <https://github.com/PaddlePaddle/paddle2onnx>`_ .


Mode deteils -> For more details,

ZeyuChen · 2020-11-11T12:13:45Z

python/paddle/onnx/export.py

+
+def export(layer, path, input_spec=None, opset_version=9, **configs):
+    """
+    Export Layer to ONNX format, which can be inferenced by onnxruntime or other backends.


which can be inferenced -> which can use for inference via onnxruntime or other backends.

ZeyuChen · 2020-11-11T12:15:00Z

python/paddle/onnx/export.py

+                    else:
+                        return y
+
+            # Export model with 'Tensor' to support prune model by set 'output_spec'.


pruned model

ZeyuChen

LGTM

TCChenlong

LGTM
TODO：review doc

XiaoguangHu01

LGTM

paddle-bot-old bot referenced this pull request Oct 12, 2020

add onnx export module, test=develop

4b9fbde

Channingss changed the title ~~add onnx export module, test=develop~~ [WIP] add onnx export module, test=develop Oct 12, 2020

ZeyuChen requested changes Oct 13, 2020

View reviewed changes

add onnx export module, test=develop

778f3fc

Channingss force-pushed the onnx branch from 4b9fbde to 48c4f06 Compare November 9, 2020 06:30

Channingss force-pushed the onnx branch from 48c4f06 to 7c2e0e5 Compare November 9, 2020 11:02

Channingss force-pushed the onnx branch from 7c2e0e5 to c755474 Compare November 10, 2020 03:11

add onnx export module, test=develop

0750280

Channingss force-pushed the onnx branch from c755474 to 0750280 Compare November 10, 2020 03:47

Channingss added 2 commits November 10, 2020 09:26

Merge branch 'develop' of https://github.com/PaddlePaddle/Paddle into…

04d75da

… onnx

add unit test for paddle.onnx.export

174e37e

Channingss force-pushed the onnx branch 3 times, most recently from 811405c to 8d7a901 Compare November 11, 2020 08:15

Channingss force-pushed the onnx branch from 8d7a901 to 07ec0e2 Compare November 11, 2020 08:34

ZeyuChen requested changes Nov 11, 2020

View reviewed changes

adjust api & doc

f9e5abd

Channingss force-pushed the onnx branch from 07ec0e2 to f9e5abd Compare November 11, 2020 11:58

ZeyuChen requested changes Nov 11, 2020

View reviewed changes

fix some typo

5cfb839

Channingss changed the title ~~[WIP] add onnx export module, test=develop~~ add onnx export module, test=develop Nov 12, 2020

ZeyuChen requested review from TCChenlong and XiaoguangHu01 November 12, 2020 06:41

Channingss requested a review from ZeyuChen November 12, 2020 06:43

ZeyuChen approved these changes Nov 12, 2020

View reviewed changes

TCChenlong approved these changes Nov 12, 2020

View reviewed changes

XiaoguangHu01 approved these changes Nov 13, 2020

View reviewed changes

ZeyuChen merged commit c545b9b into PaddlePaddle:develop Nov 13, 2020

		__all__ = ['export']


		def export(layer, save_file, input_spec=None, opset_version=9, **kwargs):

add onnx export module, test=develop #27831

add onnx export module, test=develop #27831

Conversation

Channingss commented Oct 12, 2020 • edited Loading

PR types

PR changes

Describe

paddle-bot-old bot commented Oct 12, 2020

ZeyuChen left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

ZeyuChen left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

ZeyuChen left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

ZeyuChen left a comment

Choose a reason for hiding this comment

TCChenlong left a comment

Choose a reason for hiding this comment

XiaoguangHu01 left a comment

Choose a reason for hiding this comment

Channingss commented Oct 12, 2020 •

edited

Loading