提交 de250a5d 编写于 作者: S shanyi15

delete cn_release and update en release

上级 6241f965
# v0.11.0版本
## PaddlePaddle Fluid
- 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:
https://github.com/PaddlePaddle/Paddle/blob/3df78ed2a98d37f7ae6725894cc7514effd5664b/python/paddle/v2/fluid/tests/test_while_op.py#L36-L44
- 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
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。
## 新特点
* 发布 `PaddlePaddle Fluid`
* 增加了用于模型预测的C-API。
* 用Fluid API实现了一个简单的GAN的例子。
* 增加了关于性能调优的文档。
*`paddle.v2.dataset`下载数据集提供了重试机制.
* C++中使用protobuf-lite替换protobuf减少了二进制的大小。
* 发布了新特性 [Elastic Deep Learning (EDL)](https://github.com/PaddlePaddle/cloud/tree/develop/doc/autoscale/experiment).
* 基于Bazel API利用cmake实现了一个的新的构建系统函数库。
* 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.
* [Intel® MKL-DNN on PaddlePaddle](https://github.com/PaddlePaddle/Paddle/tree/develop/doc/design/mkldnn):
- 完成了 11个 MKL-DNN 层: Convolution, Fully connectivity, Pooling, ReLU, Tanh, ELU, Softmax, BatchNorm, AddTo, Concat, LRN。
- 完成了 3个 MKL-DNN 网络: VGG-19, ResNet-50, GoogleNet
- 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。
* 增加 [softsign activation](http://www.paddlepaddle.org/docs/develop/documentation/zh/api/v2/config/activation.html#softsign)
* 增加 [dot product layer](http://www.paddlepaddle.org/docs/develop/documentation/zh/api/v2/config/layer.html#dot-prod)
* 增加 [L2 distance layer](http://www.paddlepaddle.org/docs/develop/documentation/zh/api/v2/config/layer.html#l2-distance)
* 增加 [sub-nested sequence layer](http://www.paddlepaddle.org/docs/develop/documentation/zh/api/v2/config/layer.html#sub-nested-seq)
* 增加 [kmax sequence score layer](http://www.paddlepaddle.org/docs/develop/documentation/zh/api/v2/config/layer.html#kmax-sequence-score)
* 增加 [sequence slice layer](http://www.paddlepaddle.org/docs/develop/documentation/zh/api/v2/config/layer.html#seq-slice)
* 增加 [row convolution layer](http://www.paddlepaddle.org/docs/develop/documentation/zh/api/v2/config/layer.html#row-conv)
* 增加移动端友好的网页
## 改进
* 使用一个Python`whl`包即可安装.
* [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)
*`PADDLE_ONLY_CPU` 改为 `PADDLE_WITH_GPU`, 因为我们会支持多种设备。
* 删除了有一些bug的BarrierStat。
* 清理和删除了paddle::Parameter中未使用的函数。
* 删除了ProtoDataProvider。
* Huber loss同时支持回归和分类。
* 为sequence pooling 层增加`stride`参数。
* v2 API自动使用cudnn batch normalization。
* 可以使用一个固定的参数名共享BN层的参数。
* 2D convolution operation支持variable-dimension input特性。
* 重构cmake中关于CUDA的部分并实现自动检测GPU架构的功能。
* 优化网页导航。
## 错误修复
* 修复ROI pooling的Bug. cc9a761
* 修复当label是dense vector是AUC变成0的问题. #5274
* 修复WarpCTC 层的Bug.
# v0.10.0版本
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/)
- 旧的Python API由于难以学习和使用已经过时了。使用旧版本的API至少需要两份python文件,分别是定义数据生成器和定义网络拓扑结构的文件。用户通过运行`paddle_trainer`的C++程序来启动PaddlePaddle任务,该程序调用Python解释器来运行定义网络拓扑结构的文件,然后通过迭代加载数据生成器提供的小批量数据启动训练循环。这与Python的现代编辑方式不符,比如Jupyter Notebook。
- 新版的API被称为 *V2 API*,允许我们在单个.py文件中,通过编辑更短的Python程序来定义网络结构和数据。此外,该Python程序也可以在Jupyter Notebook中运行,因为PaddlePaddle可以作为共享库来被Python程序加载和使用。
基于新的API,我们提供了一个在线的学习文档 [Deep Learning 101](http://book.paddlepaddle.org/index.en.html) 及其[中文版本](http://book.paddlepaddle.org/)
我们还致力于迭代更新新版API的在线文档,并将新版API引入分布式集群(包括MPI和Kubernetes)训练中。我们将在下一个版本中发布更多的内容。
## 新特点
* 发布新版[Python API](http://research.baidu.com/paddlepaddles-new-api-simplifies-deep-learning-programs/)
* 发布深度学习系列课程 [Deep Learning 101](http://book.paddlepaddle.org/index.en.html) 及其[中文版本](http://book.paddlepaddle.org/)
* 支持矩形输入的CNN。
* 为seqlastin和seqfirstin提供stride pooling。
*`trainer_config_helpers`中暴露`seq_concat_layer/seq_reshape_layer`
* 添加公共数据集包:CIFAR,MNIST,IMDB,WMT14,CONLL05,movielens,imikolov。
* 针对Single Shot Multibox Detection增加 Prior box layer。
* 增加光滑的L1损失。
* 在V2 API中增加 data reader 创建器和修饰器。
* 增加cmrnorm投影的CPU实现。
## 改进
* 提供`paddle_trainer`的Python virtualenv支持。
* 增加代码自动格式化的pre-commit hooks。
* 升级protobuf到3.x版本。
* 在Python数据生成器中提供一个检测数据类型的选项。
* 加速GPU中average层的后向反馈计算。
* 细化文档。
* 使用Travis-CI检查文档中的死链接。
* 增加解释`sparse_vector`的示例。
* 在layer_math.py中添加ReLU。
* 简化Quick Start示例中的数据处理流程。
* 支持CUDNN Deconv。
* 在v2 API中增加数据feeder。
* 在情感分析示例的演示中增加对标准输入流中样本的预测。
* 提供图像预处理的多进程接口。
* 增加V1 API的基准文档。
*`layer_math.py`中增加ReLU。
* 提供公共数据集的自动下载包。
*`Argument::sumCost`重新命名为`Argument::sum`,并暴露给python。
* 为矩阵相关的表达式评估增加一个新的`TensorExpression`实现。
* 增加延迟分配来优化批处理多表达式计算。
* 增加抽象的类函数及其实现:
* `PadFunc``PadGradFunc`
* `ContextProjectionForwardFunc``ContextProjectionBackwardFunc`
* `CosSimBackward``CosSimBackwardFunc`
* `CrossMapNormalFunc``CrossMapNormalGradFunc`
* `MulFunc`
* 增加`AutoCompare``FunctionCompare`类,使得编写比较gpu和cpu版本函数的单元测试更容易。
* 生成`libpaddle_test_main.a`并删除测试文件内的主函数。
* 支持PyDataProvider2中numpy的稠密向量。
* 清理代码库,删除一些复制粘贴的代码片段:
* 增加`SparseRowMatrix`的抽样类`RowBuffer`
* 清理`GradientMachine`的接口。
* 在layer中增加`override`关键字。
* 简化`Evaluator::create`,使用`ClassRegister`来创建`Evaluator`
* 下载演示的数据集时检查MD5校验。
* 添加`paddle::Error`,用于替代Paddle中的`LOG(FATAL)`
## 错误修复
* 检查`recurrent_group`的layer输入类型。
* 不要用.cu源文件运行`clang-format`
* 修复`LogActivation`的使用错误。
* 修复运行`test_layerHelpers`多次的错误。
* 修复seq2seq示例超出消息大小限制的错误。
* 修复在GPU模式下dataprovider转换的错误。
* 修复`GatedRecurrentLayer`中的错误。
* 修复在测试多个模型时`BatchNorm`的错误。
* 修复paramRelu在单元测试时崩溃的错误。
* 修复`CpuSparseMatrix`编译时相关的警告。
* 修复`MultiGradientMachine``trainer_count > batch_size`时的错误。
* 修复`PyDataProvider2`阻止异步加载数据的错误。
# Release v0.11.0
# Release Note
## PaddlePaddle Fluid
- Release 0.11.0 includes a new feature *PaddlePaddle Fluid*. Fluid is
designed to allow users to program like PyTorch and TensorFlow Eager Execution.
In these systems, there is no longer the concept *model* and applications
do not include a symbolic description of a graph of operators nor a sequence
of layers. Instead, applications look exactly like a usual program that
describes a process of training or inference. The difference between
Fluid and PyTorch or Eager Execution is that Fluid doesn't rely on Python's
control-flow, `if-then-else` nor `for`. Instead, Fluid provides its
C++ implementations and their Python binding using the `with` statement. For an example
https://github.com/PaddlePaddle/Paddle/blob/3df78ed2a98d37f7ae6725894cc7514effd5664b/python/paddle/v2/fluid/tests/test_while_op.py#L36-L44
- In 0.11.0, we provides a C++ class `Executor` to run a Fluid program.
Executor works like an interpreter. In future version, we will improve
`Executor` into a debugger like GDB, and we might provide some compilers,
which, for example, takes an application like the above one, and outputs
an equivalent C++ source program, which can be compiled using
[`nvcc`](http://docs.nvidia.com/cuda/cuda-compiler-driver-nvcc/index.html)
to generate binaries that use CUDA, or using
[`icc`](https://software.intel.com/en-us/c-compilers) to generate binaries
that make full use of Intel CPUs.
## New Features
* Release `PaddlePaddle Fluid`.
* Add C-API for model inference
* Use fluid API to create a simple GAN demo.
* Add develop guide about performance tunning.
* Add retry when download `paddle.v2.dataset`.
* Linking protobuf-lite not protobuf in C++. Reduce the binary size.
* Feature [Elastic Deep Learning (EDL)](https://github.com/PaddlePaddle/cloud/tree/develop/doc/autoscale/experiment) released.
* A new style cmake functions for Paddle. It is based on Bazel API.
* 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`.
* [Intel® MKL-DNN on PaddlePaddle](https://github.com/PaddlePaddle/Paddle/tree/develop/doc/design/mkldnn):
- Complete 11 MKL-DNN layers: Convolution, Fully connectivity, Pooling, ReLU, Tanh, ELU, Softmax, BatchNorm, AddTo, Concat, LRN.
- Complete 3 MKL-DNN networks: VGG-19, ResNet-50, GoogleNet
- [Benchmark](https://github.com/PaddlePaddle/Paddle/blob/develop/benchmark/IntelOptimizedPaddle.md) on Intel Skylake 6148 CPU: 2~3x training speedup compared with MKLML.
* Add the [`softsign` activation](http://www.paddlepaddle.org/docs/develop/documentation/zh/api/v2/config/activation.html#softsign).
* Add the [dot product layer](http://www.paddlepaddle.org/docs/develop/documentation/zh/api/v2/config/layer.html#dot-prod).
* Add the [L2 distance layer](http://www.paddlepaddle.org/docs/develop/documentation/zh/api/v2/config/layer.html#l2-distance).
* Add the [sub-nested sequence layer](http://www.paddlepaddle.org/docs/develop/documentation/zh/api/v2/config/layer.html#sub-nested-seq).
* Add the [kmax sequence score layer](http://www.paddlepaddle.org/docs/develop/documentation/zh/api/v2/config/layer.html#kmax-sequence-score).
* Add the [sequence slice layer](http://www.paddlepaddle.org/docs/develop/documentation/zh/api/v2/config/layer.html#seq-slice).
* Add the [row convolution layer](http://www.paddlepaddle.org/docs/develop/documentation/zh/api/v2/config/layer.html#row-conv)
* Add mobile friendly webpages.
## Improvements
* Build and install using a single `whl` package.
* [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).
* Change `PADDLE_ONLY_CPU` to `PADDLE_WITH_GPU`, since we will support many kinds of devices.
* Remove buggy BarrierStat.
* Clean and remove unused functions in paddle::Parameter.
* Remove ProtoDataProvider.
* Huber loss supports both regression and classification.
* Add the `stride` parameter for sequence pooling layers.
* Enable v2 API use cudnn batch normalization automatically.
* The BN layer's parameter can be shared by a fixed the parameter name.
* Support variable-dimension input feature for 2D convolution operation.
* Refine cmake about CUDA to automatically detect GPU architecture.
* Improved website navigation.
## Bug Fixes
* Fix bug in ROI pooling. cc9a761
* Fix AUC is zero when label is dense vector. #5274
* Fix bug in WarpCTC layer.
# Release v0.10.0
We are glad to release version 0.10.0. In this version, we are happy to release the new
[Python API](http://research.baidu.com/paddlepaddles-new-api-simplifies-deep-learning-programs/).
- Our old Python API is kind of out of date. It's hard to learn and hard to
use. To write a PaddlePaddle program using the old API, we'd have to write
at least two Python files: one `data provider` and another one that defines
the network topology. Users start a PaddlePaddle job by running the
`paddle_trainer` C++ program, which calls Python interpreter to run the
network topology configuration script and then start the training loop,
which iteratively calls the data provider function to load minibatches.
This prevents us from writing a Python program in a modern way, e.g., in the
Jupyter Notebook.
- The new API, which we often refer to as the *v2 API*, allows us to write
much shorter Python programs to define the network and the data in a single
.py file. Also, this program can run in Jupyter Notebook, since the entry
point is in Python program and PaddlePaddle runs as a shared library loaded
and invoked by this Python program.
Basing on the new API, we delivered an online interative
book, [Deep Learning 101](http://book.paddlepaddle.org/index.en.html)
and [its Chinese version](http://book.paddlepaddle.org/).
We also worked on updating our online documentation to describe the new API.
But this is an ongoing work. We will release more documentation improvements
in the next version.
We also worked on bring the new API to distributed model training (via MPI and
Kubernetes). This work is ongoing. We will release more about it in the next
version.
## New Features
* We release [new Python API](http://research.baidu.com/paddlepaddles-new-api-simplifies-deep-learning-programs/).
* Deep Learning 101 book in [English](http://book.paddlepaddle.org/index.en.html) and [Chinese](http://book.paddlepaddle.org/).
* Support rectangle input for CNN.
* Support stride pooling for seqlastin and seqfirstin.
* Expose `seq_concat_layer/seq_reshape_layer` in `trainer_config_helpers`.
* Add dataset package: CIFAR, MNIST, IMDB, WMT14, CONLL05, movielens, imikolov.
* Add Priorbox layer for Single Shot Multibox Detection.
* Add smooth L1 cost.
* Add data reader creator and data reader decorator for v2 API.
* Add the CPU implementation of cmrnorm projection.
## Improvements
* Support Python virtualenv for `paddle_trainer`.
* Add pre-commit hooks, used for automatically format our code.
* Upgrade protobuf to version 3.x.
* Add an option to check data type in Python data provider.
* Speedup the backward of average layer on GPU.
* Documentation refinement.
* Check dead links in documents using Travis-CI.
* Add a example for explaining `sparse_vector`.
* Add ReLU in layer_math.py
* Simplify data processing flow for Quick Start.
* Support CUDNN Deconv.
* Add data feeder in v2 API.
* Support predicting the samples from sys.stdin for sentiment demo.
* Provide multi-proccess interface for image preprocessing.
* Add benchmark document for v1 API.
* Add ReLU in `layer_math.py`.
* Add packages for automatically downloading public datasets.
* Rename `Argument::sumCost` to `Argument::sum` since class `Argument` is nothing with cost.
* Expose Argument::sum to Python
* Add a new `TensorExpression` implementation for matrix-related expression evaluations.
* Add lazy assignment for optimizing the calculation of a batch of multiple expressions.
* Add abstract calss `Function` and its implementation:
* `PadFunc` and `PadGradFunc`.
* `ContextProjectionForwardFunc` and `ContextProjectionBackwardFunc`.
* `CosSimBackward` and `CosSimBackwardFunc`.
* `CrossMapNormalFunc` and `CrossMapNormalGradFunc`.
* `MulFunc`.
* Add class `AutoCompare` and `FunctionCompare`, which make it easier to write unit tests for comparing gpu and cpu version of a function.
* Generate `libpaddle_test_main.a` and remove the main function inside the test file.
* Support dense numpy vector in PyDataProvider2.
* Clean code base, remove some copy-n-pasted code snippets:
* Extract `RowBuffer` class for `SparseRowMatrix`.
* Clean the interface of `GradientMachine`.
* Use `override` keyword in layer.
* Simplify `Evaluator::create`, use `ClassRegister` to create `Evaluator`s.
* Check MD5 checksum when downloading demo's dataset.
* Add `paddle::Error` which intentially replace `LOG(FATAL)` in Paddle.
## Bug Fixes
* Check layer input types for `recurrent_group`.
* Don't run `clang-format` with .cu source files.
* Fix bugs with `LogActivation`.
* Fix the bug that runs `test_layerHelpers` multiple times.
* Fix the bug that the seq2seq demo exceeds protobuf message size limit.
* Fix the bug in dataprovider converter in GPU mode.
* Fix a bug in `GatedRecurrentLayer`.
* Fix bug for `BatchNorm` when testing more than one models.
* Fix broken unit test of paramRelu.
* Fix some compile-time warnings about `CpuSparseMatrix`.
* Fix `MultiGradientMachine` error when `trainer_count > batch_size`.
* Fix bugs that prevents from asynchronous data loading in `PyDataProvider2`.
# Release v0.9.0
## New Features:
* New Layers
* bilinear interpolation layer.
* spatial pyramid-pool layer.
* de-convolution layer.
* maxout layer.
* Support rectangle padding, stride, window and input for Pooling Operation.
* Add —job=time in trainer, which can be used to print time info without compiler option -WITH_TIMER=ON.
* Expose cost_weight/nce_layer in `trainer_config_helpers`
* Add FAQ, concepts, h-rnn docs.
* Add Bidi-LSTM and DB-LSTM to quick start demo @alvations
* Add usage track scripts.
## Improvements
* Add Travis-CI for Mac OS X. Enable swig unittest in Travis-CI. Skip Travis-CI when only docs are changed.
* Add code coverage tools.
* Refine convolution layer to speedup and reduce GPU memory.
* Speed up PyDataProvider2
* Add ubuntu deb package build scripts.
* Make Paddle use git-flow branching model.
* PServer support no parameter blocks.
## Bug Fixes
* add zlib link to py_paddle
* add input sparse data check for sparse layer at runtime
* Bug fix for sparse matrix multiplication
* Fix floating-point overflow problem of tanh
* Fix some nvcc compile options
* Fix a bug in yield dictionary in DataProvider
* Fix SRL hang when exit.
# Release v0.8.0beta.1
New features:
* Mac OSX is supported by source code. #138
* Both GPU and CPU versions of PaddlePaddle are supported.
* Support CUDA 8.0
* Enhance `PyDataProvider2`
* Add dictionary yield format. `PyDataProvider2` can yield a dictionary with key is data_layer's name, value is features.
* Add `min_pool_size` to control memory pool in provider.
* Add `deb` install package & docker image for no_avx machines.
* Especially for cloud computing and virtual machines
* Automatically disable `avx` instructions in cmake when machine's CPU don't support `avx` instructions.
* Add Parallel NN api in trainer_config_helpers.
* Add `travis ci` for Github
Bug fixes:
* Several bugs in trainer_config_helpers. Also complete the unittest for trainer_config_helpers
* Check if PaddlePaddle is installed when unittest.
* Fix bugs in GTX series GPU
* Fix bug in MultinomialSampler
Also more documentation was written since last release.
# Release v0.8.0beta.0
PaddlePaddle v0.8.0beta.0 release. The install package is not stable yet and it's a pre-release version.
Please turn to [here](https://github.com/PaddlePaddle/Paddle/releases) for release note.
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册