Skip to content

Commit 0074046

Browse files
jacquesqiaojacquesqiao
authored
Merge pull request #6428 from jacquesqiao/release-note-v0.11.0
add v0.11.0 release note
2 parents d6f67f2 + 5d4f9fb commit 0074046

File tree

2 files changed

+131
-0
lines changed

2 files changed

+131
-0
lines changed

‎RELEASE.cn.md

Lines changed: 59 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -1,3 +1,62 @@
1+
# v0.11.0版本
2+
3+
## Fluid Python API
4+
5+
- PaddlePaddle发布版本v0.11.0包含一个新的特性*PaddlePaddle Fluid*. Fluid 是设计用来让用户像Pytorch和Tensorflow Eager Execution一样执行程序。在这些系统中，不再有*模型*这个概念，应用也不再包含一个用于描述Operator图或者一系列层的符号描述，而是像通用程序那样描述训练或者预测的过程。而Fluid与PyTorch或Eager Execution的区别在于Fluid不依赖Python提供的控制流，例如 if-else-then或者for，而是提供了基于C++实现的控制流并暴露了对应的用with语法实现的Python接口。例如：
6+
7+
https://github.com/PaddlePaddle/Paddle/blob/3df78ed2a98d37f7ae6725894cc7514effd5664b/python/paddle/v2/fluid/tests/test_while_op.py#L36-L44
8+
9+
- 在v0.11.0版本中，我们提供了一个C++类`Executor`用于运行一个Fluid程序。Executor类似一个解释器。在未来的版本中，我们将提升和优化Executor成为一个调试器，就像GDB。并可能提供一些编译器，这个编译器会读取一个上文所描述的应用然后编译成一个等价的
10+
源代码，这个源代码可以被nvcc编译成可以使用CUDA的二进制，或者被icc编译成可以充分利用Intel CPU的二进制。
11+
12+
13+
## 新特点
14+
15+
* 发布 `Fluid` API。
16+
* 增加了用于模型预测的C-API。
17+
* 用Fluid API实现了一个简单的GAN的例子。
18+
* 增加了关于性能调优的文档。
19+
* 为`paddle.v2.dataset`下载数据集提供了重试机制.
20+
* C++中使用protobuf-lite替换protobuf减少了二进制的大小。
21+
* 发布了新特性 [Elastic Deep Learning (EDL)](https://github.com/PaddlePaddle/cloud/tree/develop/doc/autoscale/experiment).
22+
* 基于Bazel API利用cmake实现了一个的新的构建系统函数库。
23+
* 当使用编译选项`WITH_MKL=ON`时自动下载和编译Intel® [MKLML](https://github.com/01org/mkl-dnn/releases/download/v0.11/mklml_lnx_2018.0.1.20171007.tgz) 函数库.
24+
* [Intel® MKL-DNN on PaddlePaddle](https://github.com/PaddlePaddle/Paddle/tree/develop/doc/design/mkldnn):
25+
- 完成了 11个 MKL-DNN 层: Convolution, Fully connectivity, Pooling, ReLU, Tanh, ELU, Softmax, BatchNorm, AddTo, Concat, LRN。
26+
- 完成了 3个 MKL-DNN 网络: VGG-19, ResNet-50, GoogleNet
27+
- 基于Intel Skylake 6148 CPU的[性能测试](https://github.com/PaddlePaddle/Paddle/blob/develop/benchmark/IntelOptimizedPaddle.md) : 相对于MKLML有2~3倍的训练加速。
28+
* 增加 [softsign activation](http://www.paddlepaddle.org/docs/develop/documentation/zh/api/v2/config/activation.html#softsign)
29+
* 增加 [dot product layer](http://www.paddlepaddle.org/docs/develop/documentation/zh/api/v2/config/layer.html#dot-prod)
30+
* 增加 [L2 distance layer](http://www.paddlepaddle.org/docs/develop/documentation/zh/api/v2/config/layer.html#l2-distance)
31+
* 增加 [sub-nested sequence layer](http://www.paddlepaddle.org/docs/develop/documentation/zh/api/v2/config/layer.html#sub-nested-seq)
32+
* 增加 [kmax sequence score layer](http://www.paddlepaddle.org/docs/develop/documentation/zh/api/v2/config/layer.html#kmax-sequence-score)
33+
* 增加 [sequence slice layer](http://www.paddlepaddle.org/docs/develop/documentation/zh/api/v2/config/layer.html#seq-slice)
34+
* 增加 [row convolution layer](http://www.paddlepaddle.org/docs/develop/documentation/zh/api/v2/config/layer.html#row-conv)
35+
* 增加移动端友好的网页
36+
37+
## 改进
38+
39+
* 使用一个Python`whl`包即可安装.
40+
* [V2 API可以实现用户定制化评估](https://github.com/PaddlePaddle/models/tree/develop/ltr#训练过程中输出自定义评估指标)。
41+
* 将 `PADDLE_ONLY_CPU` 改为 `PADDLE_WITH_GPU`, 因为我们会支持多种设备。
42+
* 删除了有一些bug的BarrierStat。
43+
* 清理和删除了paddle::Parameter中未使用的函数。
44+
* 删除了ProtoDataProvider。
45+
* Huber loss同时支持回归和分类。
46+
* 为sequence pooling 层增加`stride`参数。
47+
* v2 API自动使用cudnn batch normalization。
48+
* 可以使用一个固定的参数名共享BN层的参数。
49+
* 2D convolution operation支持variable-dimension input特性。
50+
* 重构cmake中关于CUDA的部分并实现自动检测GPU架构的功能。
51+
* 优化网页导航。
52+
53+
## 错误修复
54+
55+
* 修复ROI pooling的Bug. cc9a761
56+
* 修复当label是dense vector是AUC变成0的问题. #5274
57+
* 修复WarpCTC 层的Bug.
58+
59+
160
# v0.10.0版本
261

362
我们非常高兴发布了PaddlePaddle V0.10.0版，并开发了新的[Python API](http://research.baidu.com/paddlepaddles-new-api-simplifies-deep-learning-programs/)。

‎RELEASE.md

Lines changed: 72 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -1,3 +1,75 @@
1+
# Release v0.11.0
2+
3+
## Fluid Python API
4+
5+
- Release 0.11.0 includes a new feature *PaddlePaddle Fluid*. Fluid is
6+
designed to allow users to program like PyTorch and TensorFlow Eager Execution.
7+
In these systems, there is no longer the concept *model* and applications
8+
do not include a symbolic description of a graph of operators nor a sequence
9+
of layers. Instead, applications look exactly like a usual program that
10+
describes a process of training or inference. The difference between
11+
Fluid and PyTorch or Eager Execution is that Fluid doesn't rely on Python's
12+
control-flow, `if-then-else` nor `for`. Instead, Fluid provides its
13+
C++ implementations and their Python binding using the `with` statement. For an example
14+
15+
https://github.com/PaddlePaddle/Paddle/blob/3df78ed2a98d37f7ae6725894cc7514effd5664b/python/paddle/v2/fluid/tests/test_while_op.py#L36-L44
16+
17+
- In 0.11.0, we provides a C++ class `Executor` to run a Fluid program.
18+
Executor works like an interpreter. In future version, we will improve
19+
`Executor` into a debugger like GDB, and we might provide some compilers,
20+
which, for example, takes an application like the above one, and outputs
21+
an equivalent C++ source program, which can be compiled using
22+
[`nvcc`](http://docs.nvidia.com/cuda/cuda-compiler-driver-nvcc/index.html)
23+
to generate binaries that use CUDA, or using
24+
[`icc`](https://software.intel.com/en-us/c-compilers) to generate binaries
25+
that make full use of Intel CPUs.
26+
27+
## New Features
28+
29+
* Release `Fluid` API.
30+
* Add C-API for model inference
31+
* Use fluid API to create a simple GAN demo.
32+
* Add develop guide about performance tunning.
33+
* Add retry when download `paddle.v2.dataset`.
34+
* Linking protobuf-lite not protobuf in C++. Reduce the binary size.
35+
* Feature [Elastic Deep Learning (EDL)](https://github.com/PaddlePaddle/cloud/tree/develop/doc/autoscale/experiment) released.
36+
* A new style cmake functions for Paddle. It is based on Bazel API.
37+
* Automatically download and compile with Intel® [MKLML](https://github.com/01org/mkl-dnn/releases/download/v0.11/mklml_lnx_2018.0.1.20171007.tgz) library as CBLAS when build `WITH_MKL=ON`.
38+
* [Intel® MKL-DNN on PaddlePaddle](https://github.com/PaddlePaddle/Paddle/tree/develop/doc/design/mkldnn):
39+
- Complete 11 MKL-DNN layers: Convolution, Fully connectivity, Pooling, ReLU, Tanh, ELU, Softmax, BatchNorm, AddTo, Concat, LRN.
40+
- Complete 3 MKL-DNN networks: VGG-19, ResNet-50, GoogleNet
41+
- [Benchmark](https://github.com/PaddlePaddle/Paddle/blob/develop/benchmark/IntelOptimizedPaddle.md) on Intel Skylake 6148 CPU: 2~3x training speedup compared with MKLML.
42+
* Add the [`softsign` activation](http://www.paddlepaddle.org/docs/develop/documentation/zh/api/v2/config/activation.html#softsign).
43+
* Add the [dot product layer](http://www.paddlepaddle.org/docs/develop/documentation/zh/api/v2/config/layer.html#dot-prod).
44+
* Add the [L2 distance layer](http://www.paddlepaddle.org/docs/develop/documentation/zh/api/v2/config/layer.html#l2-distance).
45+
* Add the [sub-nested sequence layer](http://www.paddlepaddle.org/docs/develop/documentation/zh/api/v2/config/layer.html#sub-nested-seq).
46+
* Add the [kmax sequence score layer](http://www.paddlepaddle.org/docs/develop/documentation/zh/api/v2/config/layer.html#kmax-sequence-score).
47+
* Add the [sequence slice layer](http://www.paddlepaddle.org/docs/develop/documentation/zh/api/v2/config/layer.html#seq-slice).
48+
* Add the [row convolution layer](http://www.paddlepaddle.org/docs/develop/documentation/zh/api/v2/config/layer.html#row-conv)
49+
* Add mobile friendly webpages.
50+
51+
## Improvements
52+
53+
* Build and install using a single `whl` package.
54+
* [Custom evaluating in V2 API](https://github.com/PaddlePaddle/models/tree/develop/ltr#训练过程中输出自定义评估指标).
55+
* Change `PADDLE_ONLY_CPU` to `PADDLE_WITH_GPU`, since we will support many kinds of devices.
56+
* Remove buggy BarrierStat.
57+
* Clean and remove unused functions in paddle::Parameter.
58+
* Remove ProtoDataProvider.
59+
* Huber loss supports both regression and classification.
60+
* Add the `stride` parameter for sequence pooling layers.
61+
* Enable v2 API use cudnn batch normalization automatically.
62+
* The BN layer's parameter can be shared by a fixed the parameter name.
63+
* Support variable-dimension input feature for 2D convolution operation.
64+
* Refine cmake about CUDA to automatically detect GPU architecture.
65+
* Improved website navigation.
66+
67+
## Bug Fixes
68+
69+
* Fix bug in ROI pooling. cc9a761
70+
* Fix AUC is zero when label is dense vector. #5274
71+
* Fix bug in WarpCTC layer.
72+
173
# Release v0.10.0
274

375
We are glad to release version 0.10.0. In this version, we are happy to release the new

0 commit comments

Comments
 (0)