diff --git a/.gitignore b/.gitignore index 351b8204100dfd71e94cb3efa2e946b44b9e4285..1512c1438e9e0b0b7b6e0c273a24b273cb652b04 100644 --- a/.gitignore +++ b/.gitignore @@ -28,3 +28,4 @@ cmake_install.cmake paddle/.timestamp python/paddlepaddle.egg-info/ paddle/pybind/pybind.h +python/paddle/v2/framework/tests/tmp/* diff --git a/.pre-commit-config.yaml b/.pre-commit-config.yaml index 83fe9af768964003130d02b7d913ad1c2102dd1d..59661c9c1da53a2ddac0127ed1827fedde811a1d 100644 --- a/.pre-commit-config.yaml +++ b/.pre-commit-config.yaml @@ -31,6 +31,3 @@ - id: go-fmt types: - go - - id: gometalinter - types: - - go diff --git a/.travis.yml b/.travis.yml index d0e2696f100e55f320e410afd6a3038db647f76f..c51e02eb79a9e53a2b8d1d663e8f0c3e0d8c3a61 100644 --- a/.travis.yml +++ b/.travis.yml @@ -30,6 +30,7 @@ addons: - automake - libtool - ccache + ssh_known_hosts: 52.76.173.135 before_install: - if [[ "$JOB" == "check_style" ]]; then sudo ln -s /usr/bin/clang-format-3.8 /usr/bin/clang-format; fi # Paddle is using protobuf 3.1 currently. Protobuf 3.2 breaks the compatibility. So we specify the python @@ -42,6 +43,14 @@ script: - | timeout 2580 paddle/scripts/travis/${JOB}.sh # 43min timeout RESULT=$?; if [ $RESULT -eq 0 ] || [ $RESULT -eq 142 ]; then true; else false; fi; + - | + if [[ "$JOB" != "build_doc" ]]; then exit 0; fi; + if [[ "$TRAVIS_PULL_REQUEST" != "false" ]]; then exit 0; fi; + if [[ "$TRAVIS_BRANCH" != "develop" && ! "$TRAVIS_BRANCH" =~ ^v[[:digit:]]+\.[[:digit:]]+(\.[[:digit:]]+)?(-\S*)?$ ]]; then exit 0; fi; + export DEPLOY_DOCS_SH=https://raw.githubusercontent.com/PaddlePaddle/PaddlePaddle.org/master/scripts/deploy/deploy_docs.sh + export DOCS_DIR=`pwd` + cd .. + curl $DEPLOY_DOCS_SH | bash -s $CONTENT_DEC_PASSWD $TRAVIS_BRANCH $DOCS_DIR $DOCS_DIR/build/doc notifications: email: on_success: change diff --git a/CMakeLists.txt b/CMakeLists.txt index 4921226ec1c90a969fa1cfc383823820500c7757..fd3582a1bca199d62d19550ffdd1efe9db520fa7 100644 --- a/CMakeLists.txt +++ b/CMakeLists.txt @@ -86,6 +86,14 @@ if(ANDROID OR IOS) "Disable MKLDNN when cross-compiling for Android and iOS" FORCE) set(WITH_MKLML OFF CACHE STRING "Disable MKLML package when cross-compiling for Android and iOS" FORCE) + + # Compile PaddlePaddle mobile inference library + if (NOT WITH_C_API) + set(WITH_C_API ON CACHE STRING + "Always compile the C_API when cross-compiling for Android and iOS" FORCE) + endif() + set(MOBILE_INFERENCE ON) + add_definitions(-DPADDLE_MOBILE_INFERENCE) endif() set(THIRD_PARTY_PATH "${CMAKE_BINARY_DIR}/third_party" CACHE STRING @@ -97,6 +105,12 @@ if (WITH_C_API AND WITH_PYTHON) "different Python interpreter from compiling.") endif() +if(MOBILE_INFERENCE) + set(THIRD_PARTY_BUILD_TYPE MinSizeRel) +else() + set(THIRD_PARTY_BUILD_TYPE Release) +endif() + ######################################################################################## include(external/mklml) # download mklml package @@ -112,7 +126,8 @@ include(external/swig) # download, build, install swig include(external/warpctc) # download, build, install warpctc include(external/any) # download libn::any include(external/eigen) # download eigen3 -include(external/pybind11) # download pybind11 +include(external/pybind11) # download pybind11 +include(external/nccl) include(cudnn) # set cudnn libraries, must before configure include(configure) # add paddle env configuration @@ -145,7 +160,7 @@ set(EXTERNAL_LIBS if(WITH_GPU) list(APPEND EXTERNAL_LIBS ${CUDA_LIBRARIES} ${CUDA_rt_LIBRARY}) if(NOT WITH_DSO) - list(APPEND EXTERNAL_LIBS ${CUDNN_LIBRARY} ${CUDA_CUBLAS_LIBRARIES} ${CUDA_curand_LIBRARY}) + list(APPEND EXTERNAL_LIBS ${CUDNN_LIBRARY} ${CUDA_CUBLAS_LIBRARIES} ${CUDA_curand_LIBRARY} ${NCCL_LIBRARY}) endif(NOT WITH_DSO) endif(WITH_GPU) @@ -160,9 +175,11 @@ endif(USE_NNPACK) add_subdirectory(proto) -# "add_subdirectory(go)" should be placed after the following loine, -# because it depends on paddle/optimizer. -add_subdirectory(paddle/optimizer) +if(NOT MOBILE_INFERENCE) + # "add_subdirectory(go)" should be placed after the following loine, + # because it depends on paddle/optimizer. + add_subdirectory(paddle/optimizer) +endif() # "add_subdirectory(paddle)" and "add_subdirectory(python)" should be # placed after this block, because they depends on it. diff --git a/CONTRIBUTING.md b/CONTRIBUTING.md index 0d4bb973ae87bb45ef4386a63c26ed62602f2cee..a60453ff4e3bba6e6cb3b3de915dd69afd3a1ec3 100644 --- a/CONTRIBUTING.md +++ b/CONTRIBUTING.md @@ -1 +1,157 @@ -./doc/howto/dev/contribute_to_paddle_en.md +# Contribute Code + +We sincerely appreciate your contribution. This document explains our workflow and work style. + +## Workflow + +PaddlePaddle uses this [Git branching model](http://nvie.com/posts/a-successful-git-branching-model/). The following steps guide usual contributions. + +1. Fork + + Our development community has been growing fastly; it doesn't make sense for everyone to write into the official repo. So, please file Pull Requests from your fork. To make a fork, just head over to the GitHub page and click the ["Fork" button](https://help.github.com/articles/fork-a-repo/). + +1. Clone + + To make a copy of your fork to your local computers, please run + + ```bash + git clone https://github.com/your-github-account/paddle + cd paddle + ``` + +1. Create the local feature branch + + For daily works like adding a new feature or fixing a bug, please open your feature branch before coding: + + ```bash + git checkout -b my-cool-stuff + ``` + +1. Commit + + Before issuing your first `git commit` command, please install [`pre-commit`](http://pre-commit.com/) by running the following commands: + + ```bash + pip install pre-commit + pre-commit install + ``` + + Our pre-commit configuration requires clang-format 3.8 for auto-formating C/C++ code and yapf for Python. + + Once installed, `pre-commit` checks the style of code and documentation in every commit. We will see something like the following when you run `git commit`: + + ``` + ➜ git commit + CRLF end-lines remover...............................(no files to check)Skipped + yapf.................................................(no files to check)Skipped + Check for added large files..............................................Passed + Check for merge conflicts................................................Passed + Check for broken symlinks................................................Passed + Detect Private Key...................................(no files to check)Skipped + Fix End of Files.....................................(no files to check)Skipped + clang-formater.......................................(no files to check)Skipped + [my-cool-stuff c703c041] add test file + 1 file changed, 0 insertions(+), 0 deletions(-) + create mode 100644 233 + ``` + +1. Build and test + + Users can build PaddlePaddle natively on Linux and Mac OS X. But to unify the building environment and to make it easy for debugging, the recommended way is [using Docker](https://github.com/PaddlePaddle/Paddle/blob/develop/doc/howto/dev/build_en.md). + +1. Keep pulling + + An experienced Git user pulls from the official repo often -- daily or even hourly, so they notice conflicts with others work early, and it's easier to resolve smaller conflicts. + + ```bash + git remote add upstream https://github.com/PaddlePaddle/Paddle + git pull upstream develop + ``` + +1. Push and file a pull request + + You can "push" your local work into your forked repo: + + ```bash + git push origin my-cool-stuff + ``` + + The push allows you to create a pull request, requesting owners of this [official repo](https://github.com/PaddlePaddle/Paddle) to pull your change into the official one. + + To create a pull request, please follow [these steps](https://help.github.com/articles/creating-a-pull-request/). + + If your change is for fixing an issue, please write ["Fixes "](https://help.github.com/articles/closing-issues-using-keywords/) in the description section of your pull request. Github would close the issue when the owners merge your pull request. + + Please remember to specify some reviewers for your pull request. If you don't know who are the right ones, please follow Github's recommendation. + + +1. Delete local and remote branches + + To keep your local workspace and your fork clean, you might want to remove merged branches: + + ```bash + git push origin :my-cool-stuff + git checkout develop + git pull upstream develop + git branch -d my-cool-stuff + ``` + +### Code Review + +- Please feel free to ping your reviewers by sending them the URL of your pull request via IM or email. Please do this after your pull request passes the CI. + +- Please answer reviewers' every comment. If you are to follow the comment, please write "Done"; please give a reason otherwise. + +- If you don't want your reviewers to get overwhelmed by email notifications, you might reply their comments by [in a batch](https://help.github.com/articles/reviewing-proposed-changes-in-a-pull-request/). + +- Reduce the unnecessary commits. Some developers commit often. It is recommended to append a sequence of small changes into one commit by running `git commit --amend` instead of `git commit`. + + +## Coding Standard + +### Code Style + +Our C/C++ code follows the [Google style guide](http://google.github.io/styleguide/cppguide.html). + +Our Python code follows the [PEP8 style guide](https://www.python.org/dev/peps/pep-0008/). + +Our build process helps to check the code style. In [`build.sh`](https://github.com/PaddlePaddle/Paddle/blob/b84e8226514b8bb4405c3c28e54aa5077193d179/paddle/scripts/docker/build.sh#L42), the entry point of our [builder Docker image](https://github.com/PaddlePaddle/Paddle/blob/b84e8226514b8bb4405c3c28e54aa5077193d179/Dockerfile#L88), the CMake argument `WITH_STYLE_CHECK` is set to `ON` by default. This flag is on + +Please install pre-commit, which automatically reformat the changes to C/C++ and Python code whenever we run `git commit`. To check the whole codebase, we can run the command `pre-commit run -a`, as in the [`check_style.sh` file](https://github.com/PaddlePaddle/Paddle/blob/b84e8226514b8bb4405c3c28e54aa5077193d179/paddle/scripts/travis/check_style.sh#L30), which is invoked by [our Travis CI configuration](https://github.com/PaddlePaddle/Paddle/blob/b84e8226514b8bb4405c3c28e54aa5077193d179/.travis.yml#L43). + +### Unit Tests + +Please remember to add related unit tests. + +- For C/C++ code, please follow [`google-test` Primer](https://github.com/google/googletest/blob/master/googletest/docs/Primer.md). + +- For Python code, please use [Python's standard `unittest` package](http://pythontesting.net/framework/unittest/unittest-introduction/). + + +### Writing Logs + +We use [glog](https://github.com/google/glog) for logging in our C/C++ code. + +For general information, please use `LOG`. For debug information, please use [`VLOG`](http://htmlpreview.github.io/?https://github.com/google/glog/blob/master/doc/glog.html#verbose). The reason is at [here](https://groups.google.com/a/chromium.org/d/msg/chromium-dev/3NDNd1KzXeY/AZKMMx37fdQJ). + +`VLOG` requires a *verbose level* parameter. For example: + +```c++ +VLOG(3) << "Operator FC is taking " << num_inputs << "inputs." +``` + +When we run a PaddlePaddle application or test, we can specify a verbose threshold. For example: + +```bash +GLOG_vmodule=buddy_allocator=2 \ +GLOG_v=10 \ +python \ +../python/paddle/v2/framework/tests/test_recurrent_op.py +``` + +This will enable VLOG messages generated by `buddy_allocator.{h,cc}` and in the verbose range of 0 to 3, so you will see above example VLOG message, which is in level 3. This suggests that we output overall messages in lower verbose levels, so they display with higher probability. When coding C++, please follow the verbose level convention as follows: + +- verbose level 1: [framework](https://github.com/PaddlePaddle/Paddle/tree/develop/paddle/framework) +- verbose level 3: [operators](https://github.com/PaddlePaddle/Paddle/tree/develop/paddle/operators) +- verbose level 5: [memory](https://github.com/PaddlePaddle/Paddle/tree/develop/paddle/memory), [platform](https://github.com/PaddlePaddle/Paddle/tree/develop/paddle/platform) +- verbose level 7: [math](https://github.com/PaddlePaddle/Paddle/tree/develop/paddle/math) diff --git a/Dockerfile b/Dockerfile index 136db772cc6a24b8084120fa6bab666bc1eda78e..150344a8116e2be9b5bab8e5fdcc9c37f4025020 100644 --- a/Dockerfile +++ b/Dockerfile @@ -22,7 +22,7 @@ COPY ./paddle/scripts/docker/root/ /root/ RUN apt-get update && \ apt-get install -y \ - git python-pip python-dev openssh-server bison \ + git python-pip python-dev openssh-server bison libnccl-dev \ wget unzip unrar tar xz-utils bzip2 gzip coreutils ntp \ curl sed grep graphviz libjpeg-dev zlib1g-dev \ python-matplotlib gcc-4.8 g++-4.8 \ diff --git a/README.md b/README.md index b9793c3eab5d40c28f01cc67ad607b97261b3235..db0fbd88b250cdc2a3cc77521cc1c2cea77c6e87 100644 --- a/README.md +++ b/README.md @@ -51,19 +51,19 @@ Please refer to our [release announcement](https://github.com/PaddlePaddle/Paddl - **Connected to Products** In addition, PaddlePaddle is also designed to be easily deployable. At Baidu, - PaddlePaddle has been deployed into products or service with a vast number + PaddlePaddle has been deployed into products and services with a vast number of users, including ad click-through rate (CTR) prediction, large-scale image classification, optical character recognition(OCR), search ranking, computer virus detection, recommendation, etc. It is widely utilized in products at - Baidu and it has achieved a significant impact. We hope you can also exploit - the capability of PaddlePaddle to make a huge impact for your product. + Baidu and it has achieved a significant impact. We hope you can also explore + the capability of PaddlePaddle to make an impact on your product. ## Installation It is recommended to check out the [Docker installation guide](http://doc.paddlepaddle.org/develop/doc/getstarted/build_and_install/docker_install_en.html) before looking into the -[build from source guide](http://doc.paddlepaddle.org/develop/doc/getstarted/build_and_install/build_from_source_en.html) +[build from source guide](http://doc.paddlepaddle.org/develop/doc/getstarted/build_and_install/build_from_source_en.html). ## Documentation @@ -72,7 +72,7 @@ We provide [English](http://doc.paddlepaddle.org/develop/doc/) and - [Deep Learning 101](http://book.paddlepaddle.org/index.html) - You might want to start from this online interactive book that can run in Jupyter Notebook. + You might want to start from this online interactive book that can run in a Jupyter Notebook. - [Distributed Training](http://doc.paddlepaddle.org/develop/doc/howto/usage/cluster/cluster_train_en.html) diff --git a/benchmark/IntelOptimizedPaddle.md b/benchmark/IntelOptimizedPaddle.md new file mode 100644 index 0000000000000000000000000000000000000000..040f5ffa41968cbf93a817faa1db86c18956341e --- /dev/null +++ b/benchmark/IntelOptimizedPaddle.md @@ -0,0 +1,48 @@ +# Benchmark + +Machine: + +- Server + - Intel(R) Xeon(R) Gold 6148 CPU @ 2.40GHz, 2 Sockets, 20 Cores per socket +- Laptop + - DELL XPS15-9560-R1745: i7-7700HQ 8G 256GSSD + - i5 MacBook Pro (Retina, 13-inch, Early 2015) +- Desktop + - i7-6700k + +System: CentOS release 6.3 (Final), Docker 1.12.1. + +PaddlePaddle: paddlepaddle/paddle:latest (TODO: will rerun after 0.11.0) + +- MKL-DNN tag v0.10 +- MKLML 2018.0.20170720 +- OpenBLAS v0.2.20 + +On each machine, we will test and compare the performance of training on single node using MKL-DNN / MKLML / OpenBLAS respectively. + +## Benchmark Model + +### Server +Test on batch size 64, 128, 256 on Intel(R) Xeon(R) Gold 6148 CPU @ 2.40GHz + +Input image size - 3 * 224 * 224, Time: images/second + +- VGG-19 + +| BatchSize | 64 | 128 | 256 | +|--------------|-------| -----| --------| +| OpenBLAS | 7.82 | 8.62 | 10.34 | +| MKLML | 11.02 | 12.86 | 15.33 | +| MKL-DNN | 27.69 | 28.8 | 29.27 | + + +chart on batch size 128 +TBD + + - ResNet + - GoogLeNet + +### Laptop +TBD +### Desktop +TBD diff --git a/benchmark/paddle/image/provider.py b/benchmark/paddle/image/provider.py index 1ac47212b5a75667e8e9d4465b33f575516e2836..4703944c8722552d56ba80a8e0663de5fb4df53d 100644 --- a/benchmark/paddle/image/provider.py +++ b/benchmark/paddle/image/provider.py @@ -22,5 +22,5 @@ def initHook(settings, height, width, color, num_class, **kwargs): def process(settings, file_list): for i in xrange(1024): img = np.random.rand(1, settings.data_size).reshape(-1, 1).flatten() - lab = random.randint(0, settings.num_class) + lab = random.randint(0, settings.num_class - 1) yield img.astype('float32'), int(lab) diff --git a/benchmark/paddle/image/resnet.py b/benchmark/paddle/image/resnet.py new file mode 100644 index 0000000000000000000000000000000000000000..6ae1857642e8df4b3859eec68a3a5227d1c4fcb3 --- /dev/null +++ b/benchmark/paddle/image/resnet.py @@ -0,0 +1,213 @@ +#!/usr/bin/env python +from paddle.trainer_config_helpers import * + +height = 224 +width = 224 +num_class = 1000 +batch_size = get_config_arg('batch_size', int, 64) +layer_num = get_config_arg("layer_num", int, 50) +is_test = get_config_arg("is_test", bool, False) + +args = {'height': height, 'width': width, 'color': True, 'num_class': num_class} +define_py_data_sources2( + "train.list", None, module="provider", obj="process", args=args) + +settings( + batch_size=batch_size, + learning_rate=0.01 / batch_size, + learning_method=MomentumOptimizer(0.9), + regularization=L2Regularization(0.0005 * batch_size)) + + +#######################Network Configuration ############# +def conv_bn_layer(name, + input, + filter_size, + num_filters, + stride, + padding, + channels=None, + active_type=ReluActivation()): + """ + A wrapper for conv layer with batch normalization layers. + Note: + conv layer has no activation. + """ + + tmp = img_conv_layer( + name=name + "_conv", + input=input, + filter_size=filter_size, + num_channels=channels, + num_filters=num_filters, + stride=stride, + padding=padding, + act=LinearActivation(), + bias_attr=False) + return batch_norm_layer( + name=name + "_bn", input=tmp, act=active_type, use_global_stats=is_test) + + +def bottleneck_block(name, input, num_filters1, num_filters2): + """ + A wrapper for bottlenect building block in ResNet. + Last conv_bn_layer has no activation. + Addto layer has activation of relu. + """ + last_name = conv_bn_layer( + name=name + '_branch2a', + input=input, + filter_size=1, + num_filters=num_filters1, + stride=1, + padding=0) + last_name = conv_bn_layer( + name=name + '_branch2b', + input=last_name, + filter_size=3, + num_filters=num_filters1, + stride=1, + padding=1) + last_name = conv_bn_layer( + name=name + '_branch2c', + input=last_name, + filter_size=1, + num_filters=num_filters2, + stride=1, + padding=0, + active_type=LinearActivation()) + + return addto_layer( + name=name + "_addto", input=[input, last_name], act=ReluActivation()) + + +def mid_projection(name, input, num_filters1, num_filters2, stride=2): + """ + A wrapper for middile projection in ResNet. + projection shortcuts are used for increasing dimensions, + and other shortcuts are identity + branch1: projection shortcuts are used for increasing + dimensions, has no activation. + branch2x: bottleneck building block, shortcuts are identity. + """ + # stride = 2 + branch1 = conv_bn_layer( + name=name + '_branch1', + input=input, + filter_size=1, + num_filters=num_filters2, + stride=stride, + padding=0, + active_type=LinearActivation()) + + last_name = conv_bn_layer( + name=name + '_branch2a', + input=input, + filter_size=1, + num_filters=num_filters1, + stride=stride, + padding=0) + last_name = conv_bn_layer( + name=name + '_branch2b', + input=last_name, + filter_size=3, + num_filters=num_filters1, + stride=1, + padding=1) + + last_name = conv_bn_layer( + name=name + '_branch2c', + input=last_name, + filter_size=1, + num_filters=num_filters2, + stride=1, + padding=0, + active_type=LinearActivation()) + + return addto_layer( + name=name + "_addto", input=[branch1, last_name], act=ReluActivation()) + + +img = data_layer(name='image', size=height * width * 3) + + +def deep_res_net(res2_num=3, res3_num=4, res4_num=6, res5_num=3): + """ + A wrapper for 50,101,152 layers of ResNet. + res2_num: number of blocks stacked in conv2_x + res3_num: number of blocks stacked in conv3_x + res4_num: number of blocks stacked in conv4_x + res5_num: number of blocks stacked in conv5_x + """ + # For ImageNet + # conv1: 112x112 + tmp = conv_bn_layer( + "conv1", + input=img, + filter_size=7, + channels=3, + num_filters=64, + stride=2, + padding=3) + tmp = img_pool_layer(name="pool1", input=tmp, pool_size=3, stride=2) + + # conv2_x: 56x56 + tmp = mid_projection( + name="res2_1", input=tmp, num_filters1=64, num_filters2=256, stride=1) + for i in xrange(2, res2_num + 1, 1): + tmp = bottleneck_block( + name="res2_" + str(i), input=tmp, num_filters1=64, num_filters2=256) + + # conv3_x: 28x28 + tmp = mid_projection( + name="res3_1", input=tmp, num_filters1=128, num_filters2=512) + for i in xrange(2, res3_num + 1, 1): + tmp = bottleneck_block( + name="res3_" + str(i), + input=tmp, + num_filters1=128, + num_filters2=512) + + # conv4_x: 14x14 + tmp = mid_projection( + name="res4_1", input=tmp, num_filters1=256, num_filters2=1024) + for i in xrange(2, res4_num + 1, 1): + tmp = bottleneck_block( + name="res4_" + str(i), + input=tmp, + num_filters1=256, + num_filters2=1024) + + # conv5_x: 7x7 + tmp = mid_projection( + name="res5_1", input=tmp, num_filters1=512, num_filters2=2048) + for i in xrange(2, res5_num + 1, 1): + tmp = bottleneck_block( + name="res5_" + str(i), + input=tmp, + num_filters1=512, + num_filters2=2048) + + tmp = img_pool_layer( + name='avgpool', + input=tmp, + pool_size=7, + stride=1, + pool_type=AvgPooling()) + + return fc_layer(input=tmp, size=num_class, act=SoftmaxActivation()) + + +if layer_num == 50: + resnet = deep_res_net(3, 4, 6, 3) +elif layer_num == 101: + resnet = deep_res_net(3, 4, 23, 3) +elif layer_num == 152: + resnet = deep_res_net(3, 8, 36, 3) +else: + print("Wrong layer number.") + +lbl = data_layer(name="label", size=num_class) +loss = cross_entropy(name='loss', input=resnet, label=lbl) +inputs(img, lbl) +outputs(loss) diff --git a/benchmark/paddle/image/run_mkldnn.sh b/benchmark/paddle/image/run_mkldnn.sh new file mode 100755 index 0000000000000000000000000000000000000000..a4527e04968cf8c8c3c31d16f50bc3e28381f6d8 --- /dev/null +++ b/benchmark/paddle/image/run_mkldnn.sh @@ -0,0 +1,49 @@ +set -e + +function train() { + unset OMP_NUM_THREADS MKL_NUM_THREADS + export OMP_DYNAMIC="FALSE" + export KMP_AFFINITY="granularity=fine,compact,0,0" + topology=$1 + layer_num=$2 + bs=$3 + use_mkldnn=$4 + if [ $4 == "True" ]; then + thread=1 + log="logs/${topology}-${layer_num}-mkldnn-${bs}.log" + elif [ $4 == "False" ]; then + thread=`nproc` + # each trainer_count use only 1 core to avoid conflict + export OMP_NUM_THREADS=1 + export MKL_NUM_THREADS=1 + log="logs/${topology}-${layer_num}-${thread}mklml-${bs}.log" + else + echo "Wrong input $3, use True or False." + exit 0 + fi + args="batch_size=${bs},layer_num=${layer_num}" + config="${topology}.py" + paddle train --job=time \ + --config=$config \ + --use_mkldnn=$use_mkldnn \ + --use_gpu=False \ + --trainer_count=$thread \ + --log_period=10 \ + --test_period=100 \ + --config_args=$args \ + 2>&1 | tee ${log} +} + +if [ ! -d "train.list" ]; then + echo " " > train.list +fi +if [ ! -d "logs" ]; then + mkdir logs +fi + +for use_mkldnn in True False; do + for batchsize in 64 128 256; do + train vgg 19 $batchsize $use_mkldnn + train resnet 50 $batchsize $use_mkldnn + done +done diff --git a/benchmark/paddle/image/vgg.py b/benchmark/paddle/image/vgg.py new file mode 100644 index 0000000000000000000000000000000000000000..420884ed8e1ae36a3f1772bfbe8323f3d0ea71e6 --- /dev/null +++ b/benchmark/paddle/image/vgg.py @@ -0,0 +1,103 @@ +#!/usr/bin/env python +from paddle.trainer_config_helpers import * + +height = 224 +width = 224 +num_class = 1000 +batch_size = get_config_arg('batch_size', int, 64) +layer_num = get_config_arg('layer_num', int, 19) + +args = {'height': height, 'width': width, 'color': True, 'num_class': num_class} +define_py_data_sources2( + "train.list", None, module="provider", obj="process", args=args) + +settings( + batch_size=batch_size, + learning_rate=0.001 / batch_size, + learning_method=MomentumOptimizer(0.9), + regularization=L2Regularization(0.0005 * batch_size)) + +img = data_layer(name='image', size=height * width * 3) + + +def vgg_network(vgg_num=3): + tmp = img_conv_group( + input=img, + num_channels=3, + conv_padding=1, + conv_num_filter=[64, 64], + conv_filter_size=3, + conv_act=ReluActivation(), + pool_size=2, + pool_stride=2, + pool_type=MaxPooling()) + + tmp = img_conv_group( + input=tmp, + conv_num_filter=[128, 128], + conv_padding=1, + conv_filter_size=3, + conv_act=ReluActivation(), + pool_stride=2, + pool_type=MaxPooling(), + pool_size=2) + + channels = [] + for i in range(vgg_num): + channels.append(256) + tmp = img_conv_group( + input=tmp, + conv_num_filter=channels, + conv_padding=1, + conv_filter_size=3, + conv_act=ReluActivation(), + pool_stride=2, + pool_type=MaxPooling(), + pool_size=2) + channels = [] + for i in range(vgg_num): + channels.append(512) + tmp = img_conv_group( + input=tmp, + conv_num_filter=channels, + conv_padding=1, + conv_filter_size=3, + conv_act=ReluActivation(), + pool_stride=2, + pool_type=MaxPooling(), + pool_size=2) + tmp = img_conv_group( + input=tmp, + conv_num_filter=channels, + conv_padding=1, + conv_filter_size=3, + conv_act=ReluActivation(), + pool_stride=2, + pool_type=MaxPooling(), + pool_size=2) + + tmp = fc_layer( + input=tmp, + size=4096, + act=ReluActivation(), + layer_attr=ExtraAttr(drop_rate=0.5)) + + tmp = fc_layer( + input=tmp, + size=4096, + act=ReluActivation(), + layer_attr=ExtraAttr(drop_rate=0.5)) + + return fc_layer(input=tmp, size=num_class, act=SoftmaxActivation()) + + +if layer_num == 16: + vgg = vgg_network(3) +elif layer_num == 19: + vgg = vgg_network(4) +else: + print("Wrong layer number.") + +lab = data_layer('label', num_class) +loss = cross_entropy(input=vgg, label=lab) +outputs(loss) diff --git a/cmake/configure.cmake b/cmake/configure.cmake index 51c3b918cc4ef4cf6c8052ccc14028a872309fcf..24ddb24399dabeec9b8e5faf36be3eb21f420111 100644 --- a/cmake/configure.cmake +++ b/cmake/configure.cmake @@ -24,6 +24,10 @@ if(WITH_DOUBLE) add_definitions(-DPADDLE_TYPE_DOUBLE) endif(WITH_DOUBLE) +if(WITH_TESTING) + add_definitions(-DPADDLE_WITH_TESTING) +endif(WITH_TESTING) + if(NOT WITH_TIMER) add_definitions(-DPADDLE_DISABLE_TIMER) endif(NOT WITH_TIMER) @@ -49,19 +53,20 @@ if(NOT WITH_GOLANG) endif(NOT WITH_GOLANG) if(NOT WITH_GPU) - add_definitions(-DPADDLE_ONLY_CPU) add_definitions(-DHPPL_STUB_FUNC) list(APPEND CMAKE_CXX_SOURCE_FILE_EXTENSIONS cu) else() + add_definitions(-DPADDLE_WITH_CUDA) + FIND_PACKAGE(CUDA REQUIRED) if(${CUDA_VERSION_MAJOR} VERSION_LESS 7) - message(FATAL_ERROR "Paddle need CUDA >= 7.0 to compile") + message(FATAL_ERROR "Paddle needs CUDA >= 7.0 to compile") endif() if(NOT CUDNN_FOUND) - message(FATAL_ERROR "Paddle need cudnn to compile") + message(FATAL_ERROR "Paddle needs cudnn to compile") endif() set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} "-Xcompiler ${SIMD_FLAG}") diff --git a/cmake/cross_compiling/ios.cmake b/cmake/cross_compiling/ios.cmake index 0b38943952f7fb9052368fe95eb31dd7592d8a47..310450f7d009dc0cdae9c0079a96445af8ec8f95 100644 --- a/cmake/cross_compiling/ios.cmake +++ b/cmake/cross_compiling/ios.cmake @@ -79,9 +79,8 @@ if(NOT DEFINED IOS_ARCH) # FIXME(liuyiqun): support "armv7;armv7s;arm64" future set(IOS_ARCH "arm64") elseif(IOS_PLATFORM STREQUAL "SIMULATOR") - set(IOS_ARCH "i386;x86_64") - elseif(IOS_PLATFORM STREQUAL "WATCHOS") - set(IOS_ARCH armv7k) + # FIXME(liuyiqun): support "i386;x86_64" future + set(IOS_ARCH "x86_64") endif() endif() set(CMAKE_OSX_ARCHITECTURES ${IOS_ARCH} CACHE string "Build architecture for iOS") diff --git a/cmake/external/eigen.cmake b/cmake/external/eigen.cmake index f7483f6be9169eb58f0148cd3a956a8c881e1fe3..96fc886a342cae38d5b804266d3af7bc909a4da2 100644 --- a/cmake/external/eigen.cmake +++ b/cmake/external/eigen.cmake @@ -8,7 +8,7 @@ ExternalProject_Add( extern_eigen3 ${EXTERNAL_PROJECT_LOG_ARGS} GIT_REPOSITORY "https://github.com/RLovelett/eigen.git" - GIT_TAG "master" + GIT_TAG 70661066beef694cadf6c304d0d07e0758825c10 PREFIX ${EIGEN_SOURCE_DIR} UPDATE_COMMAND "" CONFIGURE_COMMAND "" diff --git a/cmake/external/gflags.cmake b/cmake/external/gflags.cmake index 957f8271e4841836956b0c3f2cf3d8c88a31192a..c819eb4d70898e48eab499c666168d78262d4240 100644 --- a/cmake/external/gflags.cmake +++ b/cmake/external/gflags.cmake @@ -36,6 +36,7 @@ ExternalProject_Add( # change this back to the official Github repo once my PR is # merged. GIT_REPOSITORY "https://github.com/wangkuiyi/gflags.git" + GIT_TAG 986964c07427ecb9cdb5bd73f73ebbd40e54dadb PREFIX ${GFLAGS_SOURCES_DIR} UPDATE_COMMAND "" CMAKE_ARGS -DCMAKE_CXX_COMPILER=${CMAKE_CXX_COMPILER} @@ -45,11 +46,11 @@ ExternalProject_Add( -DCMAKE_INSTALL_PREFIX=${GFLAGS_INSTALL_DIR} -DCMAKE_POSITION_INDEPENDENT_CODE=ON -DBUILD_TESTING=OFF - -DCMAKE_BUILD_TYPE=Release + -DCMAKE_BUILD_TYPE=${THIRD_PARTY_BUILD_TYPE} ${EXTERNAL_OPTIONAL_ARGS} CMAKE_CACHE_ARGS -DCMAKE_INSTALL_PREFIX:PATH=${GFLAGS_INSTALL_DIR} -DCMAKE_POSITION_INDEPENDENT_CODE:BOOL=ON - -DCMAKE_BUILD_TYPE:STRING=Release + -DCMAKE_BUILD_TYPE:STRING=${THIRD_PARTY_BUILD_TYPE} ) ADD_LIBRARY(gflags STATIC IMPORTED GLOBAL) diff --git a/cmake/external/glog.cmake b/cmake/external/glog.cmake index b3fef738ccc0b5886bb0a32501bb7b7adade0ff1..08bdc1e1623b0d917061c7368e9b2a8f7e9517fd 100644 --- a/cmake/external/glog.cmake +++ b/cmake/external/glog.cmake @@ -31,6 +31,7 @@ ExternalProject_Add( ${EXTERNAL_PROJECT_LOG_ARGS} DEPENDS gflags GIT_REPOSITORY "https://github.com/google/glog.git" + GIT_TAG v0.3.5 PREFIX ${GLOG_SOURCES_DIR} UPDATE_COMMAND "" CMAKE_ARGS -DCMAKE_CXX_COMPILER=${CMAKE_CXX_COMPILER} @@ -43,12 +44,12 @@ ExternalProject_Add( -DWITH_GFLAGS=ON -Dgflags_DIR=${GFLAGS_INSTALL_DIR}/lib/cmake/gflags -DBUILD_TESTING=OFF - -DCMAKE_BUILD_TYPE=Release + -DCMAKE_BUILD_TYPE=${THIRD_PARTY_BUILD_TYPE} ${EXTERNAL_OPTIONAL_ARGS} CMAKE_CACHE_ARGS -DCMAKE_INSTALL_PREFIX:PATH=${GLOG_INSTALL_DIR} -DCMAKE_INSTALL_LIBDIR:PATH=${GLOG_INSTALL_DIR}/lib -DCMAKE_POSITION_INDEPENDENT_CODE:BOOL=ON - -DCMAKE_BUILD_TYPE:STRING=Release + -DCMAKE_BUILD_TYPE:STRING=${THIRD_PARTY_BUILD_TYPE} ) ADD_LIBRARY(glog STATIC IMPORTED GLOBAL) diff --git a/cmake/external/gtest.cmake b/cmake/external/gtest.cmake index 6a2a79b7631b32e8a099797de509af64533bbb95..5a4aa7a5b71a4fdfd556a46037e6d1846d668fc4 100644 --- a/cmake/external/gtest.cmake +++ b/cmake/external/gtest.cmake @@ -56,11 +56,11 @@ IF(WITH_TESTING) -DBUILD_GMOCK=ON -Dgtest_disable_pthreads=ON -Dgtest_force_shared_crt=ON - -DCMAKE_BUILD_TYPE=Release + -DCMAKE_BUILD_TYPE=${THIRD_PARTY_BUILD_TYPE} ${EXTERNAL_OPTIONAL_ARGS} CMAKE_CACHE_ARGS -DCMAKE_INSTALL_PREFIX:PATH=${GTEST_INSTALL_DIR} -DCMAKE_POSITION_INDEPENDENT_CODE:BOOL=ON - -DCMAKE_BUILD_TYPE:STRING=Release + -DCMAKE_BUILD_TYPE:STRING=${THIRD_PARTY_BUILD_TYPE} ) ADD_LIBRARY(gtest STATIC IMPORTED GLOBAL) diff --git a/cmake/external/nccl.cmake b/cmake/external/nccl.cmake new file mode 100644 index 0000000000000000000000000000000000000000..fc43766efafc3d3e16f2906ce7f9a3d692c8e4ff --- /dev/null +++ b/cmake/external/nccl.cmake @@ -0,0 +1,67 @@ +# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve. +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + +if(NOT WITH_GPU) + return() +endif() + +include(ExternalProject) + +set(NCCL_SOURCE_DIR ${THIRD_PARTY_PATH}/nccl) + +include_directories(${NCCL_SOURCE_DIR}/src/extern_nccl/src) + +if(WITH_DSO) + # If we use DSO, we do not build nccl, just download the dependencies + set(NCCL_BUILD_COMMAND "") + set(NCCL_INSTALL_COMMAND "") + set(NCCL_INSTALL_DIR "") +else() + # otherwise, we build nccl and link it. + set(NCCL_INSTALL_DIR ${THIRD_PARTY_PATH}/install/nccl) + # Note: cuda 8.0 is needed to make nccl + # When cuda is not installed on the system directory, need to set CUDA_HOME to your cuda root + set(NCCL_BUILD_COMMAND "make -j 8") + set(NCCL_INSTALL_COMMAND "make install PREFIX=${NCCL_INSTALL_DIR}") +endif() + +ExternalProject_Add( + extern_nccl + ${EXTERNAL_PROJECT_LOG_ARGS} + GIT_REPOSITORY "https://github.com/NVIDIA/nccl.git" + GIT_TAG "v1.3.4-1" + PREFIX "${NCCL_SOURCE_DIR}" + UPDATE_COMMAND "" + CONFIGURE_COMMAND "" + BUILD_COMMAND "${NCCL_BUILD_COMMAND}" + INSTALL_COMMAND "${NCCL_INSTALL_COMMAND}" + INSTALL_DIR "${NCCL_INSTALL_DIR}" + TEST_COMMAND "" +) + +if(WITH_DSO) + if(${CMAKE_VERSION} VERSION_LESS "3.3.0") + set(dummyfile ${CMAKE_CURRENT_BINARY_DIR}/lib_nccl_dummy.c) + file(WRITE ${dummyfile} "const char * dummy_nccl = \"${dummyfile}\";") + add_library(nccl STATIC ${dummyfile}) + else() + add_library(nccl INTERFACE) + endif() +else() + add_library(nccl STATIC IMPORTED GLOBAL) + set_property(TARGET nccl PROPERTY IMPORTED_LOCATION + ${NCCL_INSTALL_DIR}/lib/libnccl_static.a) +endif() + +add_dependencies(nccl extern_nccl) diff --git a/cmake/external/openblas.cmake b/cmake/external/openblas.cmake index 143b57a954e4e6b2bf273535ebdf0fa8e3dab768..3f86e456cfbe55fe47e5b18e755e34829ebe9930 100644 --- a/cmake/external/openblas.cmake +++ b/cmake/external/openblas.cmake @@ -1,11 +1,11 @@ # Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve. -# +# # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at -# +# # http://www.apache.org/licenses/LICENSE-2.0 -# +# # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. diff --git a/cmake/external/protobuf.cmake b/cmake/external/protobuf.cmake index 7cf7ba85cca4c248dcc74e078124c0b3815ee380..be7f6a9465970711170bd15dcecaadeaa8a55f86 100644 --- a/cmake/external/protobuf.cmake +++ b/cmake/external/protobuf.cmake @@ -191,12 +191,12 @@ FUNCTION(build_protobuf TARGET_NAME BUILD_FOR_HOST) ${OPTIONAL_ARGS} -Dprotobuf_BUILD_TESTS=OFF -DCMAKE_POSITION_INDEPENDENT_CODE=ON - -DCMAKE_BUILD_TYPE=Release + -DCMAKE_BUILD_TYPE=${THIRD_PARTY_BUILD_TYPE} -DCMAKE_INSTALL_PREFIX=${PROTOBUF_INSTALL_DIR} -DCMAKE_INSTALL_LIBDIR=lib CMAKE_CACHE_ARGS -DCMAKE_INSTALL_PREFIX:PATH=${PROTOBUF_INSTALL_DIR} - -DCMAKE_BUILD_TYPE:STRING=Release + -DCMAKE_BUILD_TYPE:STRING=${THIRD_PARTY_BUILD_TYPE} -DCMAKE_VERBOSE_MAKEFILE:BOOL=OFF -DCMAKE_POSITION_INDEPENDENT_CODE:BOOL=ON ${OPTIONAL_CACHE_ARGS} diff --git a/cmake/external/pybind11.cmake b/cmake/external/pybind11.cmake index 9391c285c7544669a5b1a078b7473d7a656c1bb4..4e87dc49d8956d1fa6dec777efc5a63c6b0f79a5 100644 --- a/cmake/external/pybind11.cmake +++ b/cmake/external/pybind11.cmake @@ -1,8 +1,26 @@ -INCLUDE(ExternalProject) +# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve. +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. -SET(PYBIND_SOURCE_DIR ${THIRD_PARTY_PATH}/pybind) +if(NOT WITH_PYTHON) + return() +endif() + +include(ExternalProject) -INCLUDE_DIRECTORIES(${PYBIND_SOURCE_DIR}/src/extern_pybind/include) +set(PYBIND_SOURCE_DIR ${THIRD_PARTY_PATH}/pybind) + +include_directories(${PYBIND_SOURCE_DIR}/src/extern_pybind/include) ExternalProject_Add( extern_pybind @@ -17,14 +35,12 @@ ExternalProject_Add( TEST_COMMAND "" ) -if (${CMAKE_VERSION} VERSION_LESS "3.3.0") +if(${CMAKE_VERSION} VERSION_LESS "3.3.0") set(dummyfile ${CMAKE_CURRENT_BINARY_DIR}/pybind_dummy.c) - file(WRITE ${dummyfile} "const char * dummy_any = \"${dummyfile}\";") + file(WRITE ${dummyfile} "const char * dummy_pybind = \"${dummyfile}\";") add_library(pybind STATIC ${dummyfile}) else() add_library(pybind INTERFACE) endif() add_dependencies(pybind extern_pybind) - -LIST(APPEND external_project_dependencies pybind) diff --git a/cmake/external/swig.cmake b/cmake/external/swig.cmake index ce088ae7eaa3355f2f9761e8c421da0d7ef89fa7..9db457c7b2d61228e5d5af6827c4cda11a20a463 100644 --- a/cmake/external/swig.cmake +++ b/cmake/external/swig.cmake @@ -1,11 +1,11 @@ # Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve. -# +# # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at -# +# # http://www.apache.org/licenses/LICENSE-2.0 -# +# # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. diff --git a/cmake/external/warpctc.cmake b/cmake/external/warpctc.cmake index bb258c7b5581fc22b44f4fe15c119f8081f4767e..8bd058222880b4df3b08da09c02f9fe7f1d0ee66 100644 --- a/cmake/external/warpctc.cmake +++ b/cmake/external/warpctc.cmake @@ -35,6 +35,7 @@ ExternalProject_Add( extern_warpctc ${EXTERNAL_PROJECT_LOG_ARGS} GIT_REPOSITORY "https://github.com/gangliao/warp-ctc.git" + GIT_TAG b63a0644654a3e0ed624c85a1767bc8193aead09 PREFIX ${WARPCTC_SOURCES_DIR} UPDATE_COMMAND "" CMAKE_ARGS -DCMAKE_CXX_COMPILER=${CMAKE_CXX_COMPILER} @@ -48,9 +49,9 @@ ExternalProject_Add( -DCMAKE_DISABLE_FIND_PACKAGE_Torch=ON -DBUILD_SHARED=ON -DCMAKE_POSITION_INDEPENDENT_CODE=ON - -DCMAKE_BUILD_TYPE=Release + -DCMAKE_BUILD_TYPE=${THIRD_PARTY_BUILD_TYPE} ${EXTERNAL_OPTIONAL_ARGS} - CMAKE_CACHE_ARGS -DCMAKE_BUILD_TYPE:STRING=Release + CMAKE_CACHE_ARGS -DCMAKE_BUILD_TYPE:STRING=${THIRD_PARTY_BUILD_TYPE} -DCMAKE_POSITION_INDEPENDENT_CODE:BOOL=ON -DCMAKE_INSTALL_PREFIX:PATH=${WARPCTC_INSTALL_DIR} ) diff --git a/cmake/external/zlib.cmake b/cmake/external/zlib.cmake index c496a52b780364f3014f8fa3dfbc944a7aa7430e..a98e069b7cd1654ddd5868560d0905eab6d9c692 100644 --- a/cmake/external/zlib.cmake +++ b/cmake/external/zlib.cmake @@ -1,11 +1,11 @@ # Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve. -# +# # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at -# +# # http://www.apache.org/licenses/LICENSE-2.0 -# +# # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. @@ -42,11 +42,11 @@ ExternalProject_Add( -DBUILD_SHARED_LIBS=OFF -DCMAKE_POSITION_INDEPENDENT_CODE=ON -DCMAKE_MACOSX_RPATH=ON - -DCMAKE_BUILD_TYPE=Release + -DCMAKE_BUILD_TYPE=${THIRD_PARTY_BUILD_TYPE} ${EXTERNAL_OPTIONAL_ARGS} CMAKE_CACHE_ARGS -DCMAKE_INSTALL_PREFIX:PATH=${ZLIB_INSTALL_DIR} -DCMAKE_POSITION_INDEPENDENT_CODE:BOOL=ON - -DCMAKE_BUILD_TYPE:STRING=Release + -DCMAKE_BUILD_TYPE:STRING=${THIRD_PARTY_BUILD_TYPE} ) LIST(APPEND external_project_dependencies zlib) diff --git a/cmake/generic.cmake b/cmake/generic.cmake index d2aab938d4636b1583062e27b73cb30f5d56b7b0..c311783aa3187678c31c27ddbbd074790ca444f3 100644 --- a/cmake/generic.cmake +++ b/cmake/generic.cmake @@ -106,22 +106,22 @@ function(merge_static_libs TARGET_NAME) endforeach() list(REMOVE_DUPLICATES libs_deps) - if(APPLE) # Use OSX's libtool to merge archives - # To produce a library we need at least one source file. - # It is created by add_custom_command below and will helps - # also help to track dependencies. - set(dummyfile ${CMAKE_CURRENT_BINARY_DIR}/${TARGET_NAME}_dummy.c) + # To produce a library we need at least one source file. + # It is created by add_custom_command below and will helps + # also help to track dependencies. + set(target_SRCS ${CMAKE_CURRENT_BINARY_DIR}/${TARGET_NAME}_dummy.c) + if(APPLE) # Use OSX's libtool to merge archives # Make the generated dummy source file depended on all static input # libs. If input lib changes,the source file is touched # which causes the desired effect (relink). - add_custom_command(OUTPUT ${dummyfile} - COMMAND ${CMAKE_COMMAND} -E touch ${dummyfile} + add_custom_command(OUTPUT ${target_SRCS} + COMMAND ${CMAKE_COMMAND} -E touch ${target_SRCS} DEPENDS ${libs}) # Generate dummy staic lib - file(WRITE ${dummyfile} "const char * dummy = \"${dummyfile}\";") - add_library(${TARGET_NAME} STATIC ${dummyfile}) + file(WRITE ${target_SRCS} "const char *dummy = \"${target_SRCS}\";") + add_library(${TARGET_NAME} STATIC ${target_SRCS}) target_link_libraries(${TARGET_NAME} ${libs_deps}) foreach(lib ${libs}) @@ -130,11 +130,14 @@ function(merge_static_libs TARGET_NAME) endforeach() add_custom_command(TARGET ${TARGET_NAME} POST_BUILD COMMAND rm "${CMAKE_CURRENT_BINARY_DIR}/lib${TARGET_NAME}.a" - COMMAND /usr/bin/libtool -static -o "${CMAKE_CURRENT_BINARY_DIR}/lib${TARGET_NAME}.a" ${libfiles}) + COMMAND /usr/bin/libtool -static -o "${CMAKE_CURRENT_BINARY_DIR}/lib${TARGET_NAME}.a" ${libfiles} + ) else() # general UNIX: use "ar" to extract objects and re-add to a common lib + set(target_DIR ${CMAKE_CURRENT_BINARY_DIR}/${TARGET_NAME}.dir) + foreach(lib ${libs}) - set(objlistfile ${lib}.objlist) # list of objects in the input library - set(objdir ${lib}.objdir) + set(objlistfile ${target_DIR}/${lib}.objlist) # list of objects in the input library + set(objdir ${target_DIR}/${lib}.objdir) add_custom_command(OUTPUT ${objdir} COMMAND ${CMAKE_COMMAND} -E make_directory ${objdir} @@ -142,31 +145,32 @@ function(merge_static_libs TARGET_NAME) add_custom_command(OUTPUT ${objlistfile} COMMAND ${CMAKE_AR} -x "$" - COMMAND ${CMAKE_AR} -t "$" > ../${objlistfile} + COMMAND ${CMAKE_AR} -t "$" > ${objlistfile} DEPENDS ${lib} ${objdir} WORKING_DIRECTORY ${objdir}) - # Empty dummy source file that goes into merged library - set(mergebase ${lib}.mergebase.c) - add_custom_command(OUTPUT ${mergebase} - COMMAND ${CMAKE_COMMAND} -E touch ${mergebase} - DEPENDS ${objlistfile}) - - list(APPEND mergebases "${mergebase}") + list(APPEND target_OBJS "${objlistfile}") endforeach() - add_library(${TARGET_NAME} STATIC ${mergebases}) + # Make the generated dummy source file depended on all static input + # libs. If input lib changes,the source file is touched + # which causes the desired effect (relink). + add_custom_command(OUTPUT ${target_SRCS} + COMMAND ${CMAKE_COMMAND} -E touch ${target_SRCS} + DEPENDS ${libs} ${target_OBJS}) + + # Generate dummy staic lib + file(WRITE ${target_SRCS} "const char *dummy = \"${target_SRCS}\";") + add_library(${TARGET_NAME} STATIC ${target_SRCS}) target_link_libraries(${TARGET_NAME} ${libs_deps}) # Get the file name of the generated library - set(outlibfile "$") + set(target_LIBNAME "$") - foreach(lib ${libs}) - add_custom_command(TARGET ${TARGET_NAME} POST_BUILD - COMMAND ${CMAKE_AR} cr ${outlibfile} *.o - COMMAND ${CMAKE_RANLIB} ${outlibfile} - WORKING_DIRECTORY ${lib}.objdir) - endforeach() + add_custom_command(TARGET ${TARGET_NAME} POST_BUILD + COMMAND ${CMAKE_AR} crs ${target_LIBNAME} `find ${target_DIR} -name '*.o'` + COMMAND ${CMAKE_RANLIB} ${target_LIBNAME} + WORKING_DIRECTORY ${target_DIR}) endif() endfunction(merge_static_libs) @@ -196,7 +200,7 @@ function(cc_library TARGET_NAME) add_style_check_target(${TARGET_NAME} ${cc_library_SRCS} ${cc_library_HEADERS}) else(cc_library_SRCS) - if (cc_library_DEPS) + if(cc_library_DEPS) merge_static_libs(${TARGET_NAME} ${cc_library_DEPS}) else() message(FATAL "Please specify source file or library in cc_library.") @@ -249,7 +253,7 @@ function(nv_library TARGET_NAME) foreach(source_file ${nv_library_SRCS}) string(REGEX REPLACE "\\.[^.]*$" "" source ${source_file}) if(EXISTS ${CMAKE_CURRENT_SOURCE_DIR}/${source}.h) - list(APPEND cc_library_HEADERS ${CMAKE_CURRENT_SOURCE_DIR}/${source}.h) + list(APPEND nv_library_HEADERS ${CMAKE_CURRENT_SOURCE_DIR}/${source}.h) endif() endforeach() add_style_check_target(${TARGET_NAME} ${nv_library_SRCS} ${nv_library_HEADERS}) @@ -385,13 +389,60 @@ function(go_test TARGET_NAME) WORKING_DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR}) endfunction(go_test) +# Modification of standard 'protobuf_generate_cpp()' with protobuf-lite support +# Usage: +# paddle_protobuf_generate_cpp( ) + +function(paddle_protobuf_generate_cpp SRCS HDRS) + if(NOT ARGN) + message(SEND_ERROR "Error: paddle_protobuf_generate_cpp() called without any proto files") + return() + endif() + + set(${SRCS}) + set(${HDRS}) + + if (MOBILE_INFERENCE) + set(EXTRA_FLAG "lite:") + else() + set(EXTRA_FLAG "") + endif() + + foreach(FIL ${ARGN}) + get_filename_component(ABS_FIL ${FIL} ABSOLUTE) + get_filename_component(FIL_WE ${FIL} NAME_WE) + + set(_protobuf_protoc_src "${CMAKE_CURRENT_BINARY_DIR}/${FIL_WE}.pb.cc") + set(_protobuf_protoc_hdr "${CMAKE_CURRENT_BINARY_DIR}/${FIL_WE}.pb.h") + list(APPEND ${SRCS} "${_protobuf_protoc_src}") + list(APPEND ${HDRS} "${_protobuf_protoc_hdr}") + + add_custom_command( + OUTPUT "${_protobuf_protoc_src}" + "${_protobuf_protoc_hdr}" + + COMMAND ${CMAKE_COMMAND} -E make_directory "${CMAKE_CURRENT_BINARY_DIR}" + COMMAND ${PROTOBUF_PROTOC_EXECUTABLE} + -I${CMAKE_CURRENT_SOURCE_DIR} + --cpp_out "${EXTRA_FLAG}${CMAKE_CURRENT_BINARY_DIR}" ${ABS_FIL} + DEPENDS ${ABS_FIL} protoc + COMMENT "Running C++ protocol buffer compiler on ${FIL}" + VERBATIM ) + endforeach() + + set_source_files_properties(${${SRCS}} ${${HDRS}} PROPERTIES GENERATED TRUE) + set(${SRCS} ${${SRCS}} PARENT_SCOPE) + set(${HDRS} ${${HDRS}} PARENT_SCOPE) +endfunction() + + function(proto_library TARGET_NAME) set(oneValueArgs "") set(multiValueArgs SRCS DEPS) cmake_parse_arguments(proto_library "${options}" "${oneValueArgs}" "${multiValueArgs}" ${ARGN}) set(proto_srcs) set(proto_hdrs) - protobuf_generate_cpp(proto_srcs proto_hdrs ${proto_library_SRCS}) + paddle_protobuf_generate_cpp(proto_srcs proto_hdrs ${proto_library_SRCS}) cc_library(${TARGET_NAME} SRCS ${proto_srcs} DEPS ${proto_library_DEPS} protobuf) endfunction() diff --git a/cmake/simd.cmake b/cmake/simd.cmake index 46035a908b588861607a25d3a21cf34b7b6fd4b8..53c2de332ea74b06d1bd6e5bb119cad6af27ed01 100644 --- a/cmake/simd.cmake +++ b/cmake/simd.cmake @@ -1,27 +1,28 @@ # This file is use to check all support level of AVX on your machine # so that PaddlePaddle can unleash the vectorization power of muticore. -INCLUDE(CheckCXXSourceRuns) -INCLUDE(CheckCXXSourceCompiles) +include(CheckCXXSourceRuns) +include(CheckCXXSourceCompiles) -IF(CMAKE_COMPILER_IS_GNUCC OR CMAKE_COMPILER_IS_GNUCXX OR CMAKE_CXX_COMPILER_ID MATCHES "Clang") +if(CMAKE_COMPILER_IS_GNUCC OR CMAKE_COMPILER_IS_GNUCXX OR CMAKE_CXX_COMPILER_ID MATCHES "Clang") set(MMX_FLAG "-mmmx") set(SSE2_FLAG "-msse2") set(SSE3_FLAG "-msse3") - SET(AVX_FLAG "-mavx") - SET(AVX2_FLAG "-mavx2") -ELSEIF(MSVC) + set(AVX_FLAG "-mavx") + set(AVX2_FLAG "-mavx2") +elseif(MSVC) set(MMX_FLAG "/arch:MMX") set(SSE2_FLAG "/arch:SSE2") set(SSE3_FLAG "/arch:SSE3") SET(AVX_FLAG "/arch:AVX") SET(AVX2_FLAG "/arch:AVX2") -ENDIF() +endif() set(CMAKE_REQUIRED_FLAGS_RETAINED ${CMAKE_REQUIRED_FLAGS}) # Check MMX set(CMAKE_REQUIRED_FLAGS ${MMX_FLAG}) +set(MMX_FOUND_EXITCODE 1 CACHE STRING "Result from TRY_RUN" FORCE) CHECK_CXX_SOURCE_RUNS(" #include int main() @@ -32,6 +33,7 @@ int main() # Check SSE2 set(CMAKE_REQUIRED_FLAGS ${SSE2_FLAG}) +set(SSE2_FOUND_EXITCODE 1 CACHE STRING "Result from TRY_RUN" FORCE) CHECK_CXX_SOURCE_RUNS(" #include int main() @@ -42,6 +44,7 @@ int main() # Check SSE3 set(CMAKE_REQUIRED_FLAGS ${SSE3_FLAG}) +set(SSE3_FOUND_EXITCODE 1 CACHE STRING "Result from TRY_RUN" FORCE) CHECK_CXX_SOURCE_RUNS(" #include int main() @@ -55,6 +58,7 @@ int main() # Check AVX set(CMAKE_REQUIRED_FLAGS ${AVX_FLAG}) +set(AVX_FOUND_EXITCODE 1 CACHE STRING "Result from TRY_RUN" FORCE) CHECK_CXX_SOURCE_RUNS(" #include int main() @@ -67,6 +71,7 @@ int main() # Check AVX 2 set(CMAKE_REQUIRED_FLAGS ${AVX2_FLAG}) +set(AVX2_FOUND_EXITCODE 1 CACHE STRING "Result from TRY_RUN" FORCE) CHECK_CXX_SOURCE_RUNS(" #include int main() diff --git a/cmake/util.cmake b/cmake/util.cmake index e814cad36f2a8ce95a2dc9fabc35cb39506d4cd7..117ab7f49cdf4a568cd203b2b17767643d0b2d50 100644 --- a/cmake/util.cmake +++ b/cmake/util.cmake @@ -25,7 +25,7 @@ function(target_circle_link_libraries TARGET_NAME) endif() endforeach() if("${CMAKE_CXX_COMPILER_ID}" STREQUAL "Clang" OR "${CMAKE_CXX_COMPILER_ID}" STREQUAL "AppleClang") - if(IOS AND NOT IOS_ENABLE_BITCODE) + if(NOT IOS_ENABLE_BITCODE) list(APPEND LIBS "-undefined dynamic_lookup") endif() endif() @@ -73,30 +73,52 @@ function(link_paddle_exe TARGET_NAME) generate_rdma_links() endif() - target_circle_link_libraries(${TARGET_NAME} - ARCHIVE_START - paddle_gserver - paddle_function - ARCHIVE_END - paddle_pserver - paddle_trainer_lib - paddle_network - paddle_math - paddle_utils - paddle_parameter - paddle_proto - paddle_cuda - paddle_optimizer - ${EXTERNAL_LIBS} - ${CMAKE_THREAD_LIBS_INIT} - ${CMAKE_DL_LIBS} - ${RDMA_LD_FLAGS} - ${RDMA_LIBS}) + if(MOBILE_INFERENCE) + target_circle_link_libraries(${TARGET_NAME} + ARCHIVE_START + paddle_gserver + paddle_function + ARCHIVE_END + paddle_math + paddle_utils + paddle_parameter + paddle_proto + paddle_cuda + ${EXTERNAL_LIBS} + ${CMAKE_THREAD_LIBS_INIT} + ${CMAKE_DL_LIBS} + ${RDMA_LD_FLAGS} + ${RDMA_LIBS}) + else() + target_circle_link_libraries(${TARGET_NAME} + ARCHIVE_START + paddle_gserver + paddle_function + ARCHIVE_END + paddle_pserver + paddle_trainer_lib + paddle_network + paddle_math + paddle_utils + paddle_parameter + paddle_proto + paddle_cuda + paddle_optimizer + ${EXTERNAL_LIBS} + ${CMAKE_THREAD_LIBS_INIT} + ${CMAKE_DL_LIBS} + ${RDMA_LD_FLAGS} + ${RDMA_LIBS}) + endif() if(ANDROID) target_link_libraries(${TARGET_NAME} log) endif(ANDROID) + if(WITH_MKLDNN AND WITH_MKLML AND MKLDNN_IOMP_DIR) + target_link_libraries(${TARGET_NAME} "-L${MKLDNN_IOMP_DIR} -liomp5 -Wl,--as-needed") + endif() + add_dependencies(${TARGET_NAME} ${external_project_dependencies}) endfunction() diff --git a/doc/api/v1/index_cn.rst b/doc/api/v1/index_cn.rst index 3718cd73a2003b8ef6c406a9bd51dc68e76402dc..cf146dc088e3905a751ff55c26fd82ef0ba02c89 100644 --- a/doc/api/v1/index_cn.rst +++ b/doc/api/v1/index_cn.rst @@ -21,7 +21,7 @@ Model Config API trainer_config_helpers/optimizers.rst trainer_config_helpers/data_sources.rst trainer_config_helpers/layers.rst - trainer_config_helpers/activations.rst + trainer_config_helpers/activations.rst trainer_config_helpers/poolings.rst trainer_config_helpers/networks.rst trainer_config_helpers/evaluators.rst diff --git a/doc/api/v2/config/layer.rst b/doc/api/v2/config/layer.rst index c94627a72806fa2eca77c79da24f7f3ca18f0259..d4e9d53e5c0955912a594fe8cd9cd41a4080a2d2 100644 --- a/doc/api/v2/config/layer.rst +++ b/doc/api/v2/config/layer.rst @@ -345,6 +345,11 @@ clip .. autoclass:: paddle.v2.layer.clip :noindex: +resize +------ +.. autoclass:: paddle.v2.layer.resize + :noindex: + slope_intercept --------------- .. autoclass:: paddle.v2.layer.slope_intercept diff --git a/doc/api/v2/config/networks.rst b/doc/api/v2/config/networks.rst index 6e813ab1a820d068ea3e54cad6178f1cf928eadc..048379cf01f4aec5e73e2fe3ddfa728f3c17a5d1 100644 --- a/doc/api/v2/config/networks.rst +++ b/doc/api/v2/config/networks.rst @@ -125,3 +125,8 @@ simple_attention :members: simple_attention :noindex: +dot_product_attention +--------------------- +.. automodule:: paddle.v2.networks + :members: dot_product_attention + :noindex: diff --git a/doc/design/block.md b/doc/design/block.md index be8800122035984df281692fc40009c397565046..4066122c0e8dfa33776796c3d205ba5aec9e0f52 100644 --- a/doc/design/block.md +++ b/doc/design/block.md @@ -5,12 +5,12 @@ Both deep learning systems and programming languages help users describe computation procedures. These systems use various representations of computation: - Caffe, Torch, and Paddle: sequences of layers. -- TensorFlow, Caffe2, Mxnet: graphs of operators. +- TensorFlow, Caffe2, Mxnet: graph of operators. - PaddlePaddle: nested blocks, like C++ and Java programs. ## Block in Programming Languages and Deep Learning -In programming languages, a block is a pair of curly braces that includes local variables definitions and a sequence of instructions, or operators. +In programming languages, a block is a pair of curly braces that includes local variables definitions and a sequence of instructions or operators. Blocks work with control flow structures like `if`, `else`, and `for`, which have equivalents in deep learning: @@ -24,14 +24,14 @@ A key difference is that a C++ program describes a one pass computation, whereas ## Stack Frames and the Scope Hierarchy -The existence of the backward makes the execution of a block of traditional programs and PaddlePaddle different to each other: +The existence of the backward pass makes the execution of a block of PaddlePaddle different from traditional programs: -| programming languages | PaddlePaddle | -|-----------------------|-------------------------------| -| stack | scope hierarchy | -| stack frame | scope | -| push at entering block| push at entering block | -| pop at leaving block | destroy at minibatch completes| +| programming languages | PaddlePaddle | +|-----------------------|---------------------------------| +| stack | scope hierarchy | +| stack frame | scope | +| push at entering block| push at entering block | +| pop at leaving block | destroy when minibatch completes| 1. In traditional programs: @@ -42,9 +42,9 @@ The existence of the backward makes the execution of a block of traditional prog 1. In PaddlePaddle - When the execution enters a block, PaddlePaddle adds a new scope, where it realizes variables. - - PaddlePaddle doesn't pop a scope after the execution of the block because variables therein are to be used by the backward pass. So it has a stack forest known as a *scope hierarchy*. + - PaddlePaddle doesn't pop a scope after the execution of the block because variables therein are used by the backward pass. So it has a stack forest known as a *scope hierarchy*. - The height of the highest tree is the maximum depth of nested blocks. - - After the process of a minibatch, PaddlePaddle destroys the scope hierarchy. + - After the processing of a minibatch, PaddlePaddle destroys the scope hierarchy. ## Use Blocks in C++ and PaddlePaddle Programs @@ -55,17 +55,23 @@ Let us consolidate the discussion by presenting some examples. The following C++ programs shows how blocks are used with the `if-else` structure: ```c++ +namespace pd = paddle; + int x = 10; -int y = 20; -int out; +int y = 1; +int z = 10; bool cond = false; +int o1, o2; if (cond) { int z = x + y; - out = softmax(z); + o1 = z; + o2 = pd::layer::softmax(z); } else { - int z = fc(x); - out = z; + int d = pd::layer::fc(z); + o1 = d; + o2 = d+1; } + ``` An equivalent PaddlePaddle program from the design doc of the [IfElseOp operator](./if_else_op.md) is as follows: @@ -73,57 +79,55 @@ An equivalent PaddlePaddle program from the design doc of the [IfElseOp operator ```python import paddle as pd -x = var(10) -y = var(20) -cond = var(false) -ie = pd.create_ifelseop(inputs=[x], output_num=1) +x = minibatch([10, 20, 30]) # shape=[None, 1] +y = var(1) # shape=[1], value=1 +z = minibatch([10, 20, 30]) # shape=[None, 1] +cond = larger_than(x, 15) # [false, true, true] + +ie = pd.ifelse() with ie.true_block(): - x = ie.inputs(true, 0) - z = operator.add(x, y) - ie.set_output(true, 0, operator.softmax(z)) + d = pd.layer.add_scalar(x, y) + ie.output(d, pd.layer.softmax(d)) with ie.false_block(): - x = ie.inputs(false, 0) - z = layer.fc(x) - ie.set_output(true, 0, operator.softmax(z)) -out = b(cond) + d = pd.layer.fc(z) + ie.output(d, d+1) +o1, o2 = ie(cond) ``` -In both examples, the left branch computes `softmax(x+y)` and the right branch computes `fc(x)`. +In both examples, the left branch computes `x+y` and `softmax(x+y)`, the right branch computes `fc(x)` and `x+1` . + +The difference is that variables in the C++ program contain scalar values, whereas those in the PaddlePaddle programs are mini-batches of instances. -A difference is that variables in the C++ program contain scalar values, whereas those in the PaddlePaddle programs are mini-batches of instances. The `ie.input(true, 0)` invocation returns instances in the 0-th input, `x`, that corresponds to true values in `cond` as the local variable `x`, where `ie.input(false, 0)` returns instances corresponding to false values. ### Blocks with `for` and `RNNOp` -The following RNN model from the [RNN design doc](./rnn.md) +The following RNN model in PaddlePaddle from the [RNN design doc](./rnn.md) : ```python -x = sequence([10, 20, 30]) -m = var(0) -W = tensor() -U = tensor() - -rnn = create_rnn(inputs=[input]) -with rnn.stepnet() as net: - x = net.set_inputs(0) - h = net.add_memory(init=m) - fc_out = pd.matmul(W, x) - hidden_out = pd.matmul(U, h.pre(n=1)) - sum = pd.add_two(fc_out, hidden_out) - act = pd.sigmoid(sum) - h.update(act) # update memory with act - net.set_outputs(0, act, hidden_out) # two outputs - +x = sequence([10, 20, 30]) # shape=[None, 1] +m = var(0) # shape=[1] +W = var(0.314, param=true) # shape=[1] +U = var(0.375, param=true) # shape=[1] + +rnn = pd.rnn() +with rnn.step(): + h = rnn.memory(init = m) + h_prev = rnn.previous_memory(h) + a = layer.fc(W, x) + b = layer.fc(U, h_prev) + s = pd.add(a, b) + act = pd.sigmoid(s) + rnn.update_memory(h, act) + rnn.output(a, b) o1, o2 = rnn() -print o1, o2 ``` - has its equivalent C++ program as follows ```c++ int* x = {10, 20, 30}; -int m = 0; -int W = some_value(); -int U = some_other_value(); +int* m = {0}; +int* W = {0.314}; +int* U = {0.375}; int mem[sizeof(x) / sizeof(x[0]) + 1]; int o1[sizeof(x) / sizeof(x[0]) + 1]; @@ -131,25 +135,21 @@ int o2[sizeof(x) / sizeof(x[0]) + 1]; for (int i = 1; i <= sizeof(x)/sizeof(x[0]); ++i) { int x = x[i-1]; if (i == 1) mem[0] = m; - int fc_out = W * x; - int hidden_out = Y * mem[i-1]; - int sum = fc_out + hidden_out; + int a = W * x; + int b = Y * mem[i-1]; + int s = fc_out + hidden_out; int act = sigmoid(sum); mem[i] = act; o1[i] = act; o2[i] = hidden_out; } - -print_array(o1); -print_array(o2); ``` - ## Compilation and Execution -Like TensorFlow programs, a PaddlePaddle program is written in Python. The first part describes a neural network as a protobuf message, and the rest part executes the message for training or inference. +Like TensorFlow, a PaddlePaddle program is written in Python. The first part describes a neural network as a protobuf message, and the rest executes the message for training or inference. -The generation of this protobuf message is like what a compiler generates a binary executable file. The execution of the message that the OS executes the binary file. +The generation of this protobuf message is similar to how a compiler generates a binary executable file. The execution of the message is similar to how the OS executes the binary file. ## The "Binary Executable File Format" @@ -186,10 +186,10 @@ Also, the RNN operator in above example is serialized into a protobuf message of ``` OpDesc { - inputs = {0} // the index of x - outputs = {5, 3} // indices of act and hidden_out + inputs = {0} // the index of x in vars of BlockDesc above + outputs = {5, 3} // indices of act and hidden_out in vars of BlockDesc above attrs { - "memories" : {1} // the index of h + "states" : {1} // the index of h "step_net" : } }; @@ -203,32 +203,32 @@ This `OpDesc` value is in the `ops` field of the `BlockDesc` value representing During the generation of the Protobuf message, the Block should store VarDesc (the Protobuf message which describes Variable) and OpDesc (the Protobuf message which describes Operator). VarDesc in a block should have its name scope to avoid local variables affect parent block's name scope. -Child block's name scopes should inherit the parent's so that OpDesc in child block can reference a VarDesc that stored in parent block. For example +Child block's name scopes should inherit the parent's so that OpDesc in child block can reference a VarDesc that stored in parent block. For example: ```python -a = pd.Varaible(shape=[20, 20]) +a = pd.Variable(shape=[20, 20]) b = pd.fc(a, params=["fc.w", "fc.b"]) rnn = pd.create_rnn() -with rnn.stepnet() as net: - x = net.set_inputs(a) +with rnn.stepnet(): + x = a.as_step_input() # reuse fc's parameter fc_without_b = pd.get_variable("fc.w") - net.set_outputs(fc_without_b) + rnn.output(fc_without_b) out = rnn() ``` -the method `pd.get_variable` can help retrieve a Variable by a name, a Variable may store in a parent block, but might be retrieved in a child block, so block should have a variable scope that supports inheritance. +The method `pd.get_variable` can help retrieve a Variable by the name. The Variable may be stored in a parent block, but might be retrieved in a child block, so block should have a variable scope that supports inheritance. In compiler design, the symbol table is a data structure created and maintained by compilers to store information about the occurrence of various entities such as variable names, function names, classes, etc. To store the definition of variables and operators, we define a C++ class `SymbolTable`, like the one used in compilers. -`SymbolTable` can do the following stuff: +`SymbolTable` can do the following: - store the definitions (some names and attributes) of variables and operators, -- to verify if a variable was declared, -- to make it possible to implement type checking (offer Protobuf message pointers to `InferShape` handlers). +- verify if a variable was declared, +- make it possible to implement type checking (offer Protobuf message pointers to `InferShape` handlers). ```c++ @@ -240,19 +240,18 @@ class SymbolTable { OpDesc* NewOp(const string& name=""); - // TODO determine whether name is generated by python or C++ - // currently assume that a unique name will be generated by C++ if the - // argument name left default. - VarDesc* NewVar(const string& name=""); + // TODO determine whether name is generated by python or C++. + // Currently assume that a unique name will be generated by C++ if the + // argument name is left default. + VarDesc* Var(const string& name=""); - // find a VarDesc by name, if recursive true, find parent's SymbolTable + // find a VarDesc by name, if recursive is true, find parent's SymbolTable // recursively. // this interface is introduced to support InferShape, find protobuf messages // of variables and operators, pass pointers into InferShape. - // operator // // NOTE maybe some C++ classes such as VarDescBuilder and OpDescBuilder should - // be proposed and embedded into pybind to enable python operate on C++ pointers. + // be proposed and embedded into pybind to enable python operation on C++ pointers. VarDesc* FindVar(const string& name, bool recursive=true); OpDesc* FindOp(const string& name); @@ -270,7 +269,7 @@ class SymbolTable { After all the description of variables and operators is added into SymbolTable, the block has enough information to run. -The `Block` class takes a `BlockDesc` as input, and provide `Run` and `InferShape` functions. +The `Block` class takes a `BlockDesc` as input, and provides `Run` and `InferShape` functions. ```c++ @@ -302,7 +301,7 @@ public: void CreateVariables(const framework::Scope& scope); void CreateOperators(); - // some other necessary interfaces of NetOp are list below + // some other necessary interfaces of NetOp are listed below // ... private: @@ -316,15 +315,14 @@ private: Block inherits from OperatorBase, which has a Run method. Block's Run method will run its operators sequentially. -There is another important interface called `Eval`, which take some arguments called targets, and generate a minimal graph which takes targets as the end points and creates a new Block, -after `Run`, `Eval` will get the latest value and return the targets. +There is another important interface called `Eval`, which takes some arguments called targets and generates a minimal graph which treats targets as the end points and creates a new Block. After `Run`, `Eval` will get the latest value and return the targets. The definition of Eval is as follows: ```c++ // clean a block description by targets using the corresponding dependency graph. // return a new BlockDesc with minimal number of operators. -// NOTE not return a Block but the block's description so that this can be distributed +// NOTE: The return type is not a Block but the block's description so that this can be distributed // to a cluster. BlockDesc Prune(const BlockDesc& desc, vector targets); diff --git a/doc/design/cluster_train/src/trainer.graffle b/doc/design/cluster_train/src/trainer.graffle index 42384a3f059966e22e22f5fa4295cc9ead5cef83..43415ed8cf61a5acfa34f8e56b9577f338dbf254 100644 Binary files a/doc/design/cluster_train/src/trainer.graffle and b/doc/design/cluster_train/src/trainer.graffle differ diff --git a/doc/design/dcgan.png b/doc/design/dcgan.png new file mode 100644 index 0000000000000000000000000000000000000000..15e8e290a111ff43900934341365cb4360d87d28 Binary files /dev/null and b/doc/design/dcgan.png differ diff --git a/doc/design/executor.md b/doc/design/executor.md new file mode 100644 index 0000000000000000000000000000000000000000..b5fb6c5c3c1da3c112ce63878322083dd5c42b70 --- /dev/null +++ b/doc/design/executor.md @@ -0,0 +1,23 @@ +# Executor Design Doc + +## Motivation + +We use executor to do the runtime evaluation of a `ProgramDesc`. + +## Overview + +An executor takes a `ProgramDesc`, a `block_id` and a `Scope`. The `ProgramDesc` is a list of blocks and each block contains the protobuf definition of all the parameters and operators. The `block_id` specifies the entrance block. And the `Scope` is the container of all the variable instance, which is persistent throughout different runs. + +### What does executor do? + +It evaluates all the operators in the `block_id`th block of a `ProgramDesc`. + +### What does executor NOT do? + +It does not do runtime optimization, meaning intelligently parse the dependency of each op a choose which one to be run and in which order they should be run. + +It does not do graph partitioning, meaning dividing the `ProgramDesc` into several small pieces and executing them on different devices. + +## Implementation + +`Executor` evaluates a `ProgramDesc`. Essentially, it instantiates Variables and Operators, then run all the operators in sequence. [[code]](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/framework/executor.cc) diff --git a/doc/design/float16.md b/doc/design/float16.md new file mode 100644 index 0000000000000000000000000000000000000000..078801ba2ed969d26dd31d5ec4ed268686cf7016 --- /dev/null +++ b/doc/design/float16.md @@ -0,0 +1,60 @@ +# Design Doc: float16 + +## Why float16 +Half precision (float16) is a binary floating-point format that occupies 16 bits in memory. float16 is half the size of traditional 32-bit single precision format (float) and has lower precision and smaller range. + +When high precision computation is not required, using float16 data type could potentially + +- reduce storage space, memory bandwidth, and power usages; +- increase the chance of data fitting into a smaller cache of lower latency; +- provide arithmetic speed up if supported by hardware. + +## Survey of current float16 support +A brief survey of float16 support on different compilers, hardwares, and libraries can be found below. Interested readers can refer to [link1](https://github.com/PaddlePaddle/Paddle/issues/4853) and [link2](https://github.com/Xreki/Xreki.github.io/blob/master/multi_data_types_in_dl_framework/ppt/float16_and_quantized_type.md) for more info. + +The goal of float16 is to serve as a key for the executor to find and run the correct version of compute method specialized for float16 in operator kernel. It should be compatible with various natively supported float16 implementations including `__half` for cuda, `float16_t` for ARM, and `Eigen::half` for Eigen to make writing customized float16 kernels easier. + +### Compiler +- nvcc supports `__half` data type after CUDA 7.5. +- `__fp16` or `float16_t` is supported as storage type for gcc >= 6.1 and clang >= 3.4. +- `__fp16` or `float16_t` is supported as arithmetic type for gcc >= 7.1 and clang >= 3.9. + +### Hardware +- `__half` is supported on GPU with compute capability >= 5.3. +- `__fp16` is supported as storage type for ARMv7-A, ARMv8-A, and above. +- `__fp16` is supported as arithmetic type after ARMv8.2-A (currently, the only microarchitecture implementing ARMv8.2-A is ARM Cortex-A75, which is announced in May 2017. There seems to be no application processors currently available on market that adopts this architecture. It is reported that Qualcomm Snapdragon 845 uses Cortex-A75 design and will be available in mobile devices in early 2018). + +### Libraries +- [Eigen](https://github.com/RLovelett/eigen) >= 3.3 supports float16 calculation on both GPU and CPU using the `Eigen::half` class. It is mostly useful for Nvidia GPUs because of the overloaded arithmetic operators using cuda intrinsics. It falls back to using software emulation on CPU for calculation and there is no special treatment to ARM processors. +- [ARM compute library](https://github.com/ARM-software/ComputeLibrary) >= 17.02.01 supports NEON FP16 kernels (requires ARMv8.2-A CPU). + + +## Implementation +The float16 class holds a 16-bit `uint16_t` data internally. +``` +struct float16 { + uint16_t x; +}; +``` + +float16 supports the following features: + - constructors / assignment operators that take input from primitive data types including bool, integers of various length, float, and double. + - constructors / assignment operators that take input from `__half` on cuda, `float16_t` on ARM, and `Eigen::half` on Eigen. + - conversion operators to primitive data types and half precision data types on cuda, ARM and Eigen. + - overloaded arithmetic operators for cuda, arm, and non-arm cpu, respectively. These operators will take advantage of the cuda and ARM intrinsics on the corresponding hardware. + +To support the above features, two fundamental conversion functions are provided: +``` +float16 float_to_half_rn(float f); // convert to half precision in round-to-nearest-even mode +float half_to_float(float16 h); +``` +which provides one-to-one conversion between float32 and float16. These twos functions will do different conversion routines based on the current hardware. CUDA/ARM instrinsics will be used when the corresonding hardware is available. If the hardware or compiler level does not support float32 to float16 conversion, software emulation will be performed to do the conversion. + +## To do +After float16 class is available, some of the future items are below: + +- Update pybind/tensor_py.h to bind c++ float16 with numpy float16. + +- Modify `GetKernelType()` method in `framework/operator.h` to make it compatible with float16. + +- Create a type-casting operator that can convert the data type in tensor between float16 and other types. diff --git a/doc/design/gan_api.md b/doc/design/gan_api.md new file mode 100644 index 0000000000000000000000000000000000000000..fb41df8615f73d9fd4c32995eab265833eac1a55 --- /dev/null +++ b/doc/design/gan_api.md @@ -0,0 +1,253 @@ +# Design for GAN + +GAN (General Adversarial Net [https://arxiv.org/abs/1406.2661]) is an important model for unsupervised learning and widely used in many areas. + +It applies several important concepts in machine learning system design, including building and running subgraphs, dependency tracing, different optimizers in one executor and so forth. + +In our GAN design, we wrap it as a user-friendly easily customized python API to design different models. We take the conditional DC-GAN (Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks [https://arxiv.org/abs/1511.06434]) as an example due to its good performance on image generation. + +

+
+Figure 1. The overall running logic of GAN. The black solid arrows indicate the forward pass; the green dashed arrows indicate the backward pass of generator training; the red dashed arrows indicate the backward pass of the discriminator training. The BP pass of the green (red) arrow should only update the parameters in the green (red) boxes. The diamonds indicate the data providers. d\_loss and g\_loss marked in red and green are the two targets we would like to run. +

+ +The operators, layers and functions required/optional to build a GAN demo is summarized in https://github.com/PaddlePaddle/Paddle/issues/4563. + +

+
+Figure 2. Photo borrowed from the original DC-GAN paper. +

+ +## The Conditional-GAN might be a class. +This design we adopt the popular open source design in https://github.com/carpedm20/DCGAN-tensorflow and https://github.com/rajathkmp/DCGAN. It contains following data structure: + +- DCGAN(object): which contains everything required to build a GAN model. It provides following member functions methods as API: + +- __init__(...): Initialize hyper-parameters (like conv dimension and so forth), and declare model parameters of discriminator and generator as well. + +- generator(z, y=None): Generate a fake image from input noise z. If the label y is provided, the conditional GAN model will be chosen. +Returns a generated image. + +- discriminator(image): +Given an image, decide if it is from a real source or a fake one. +Returns a 0/1 binary label. + +- build_model(self): +build the whole GAN model, define training loss for both generator and discrimator. + +## Discussion on Engine Functions required to build GAN +- Trace the tensor and variable dependency in the engine executor. (Very critical, otherwise GAN can'be be trained correctly) +- Different optimizers responsible for optimizing different loss. + +To be more detailed, we introduce our design of DCGAN as following: + +### Class member Function: Initializer +- Set up hyper-parameters, including condtional dimension, noise dimension, batch size and so forth. +- Declare and define all the model variables. All the discriminator parameters are included in the list self.theta_D and all the generator parameters are included in the list self.theta_G. +```python +class DCGAN(object): + def __init__(self, y_dim=None): + + # hyper parameters + self.y_dim = y_dim # conditional gan or not + self.batch_size = 100 + self.z_dim = z_dim # input noise dimension + + # define parameters of discriminators + self.D_W0 = pd.Variable(shape=[3,3, 1, 128], data=pd.gaussian_normal_randomizer()) + self.D_b0 = pd.Variable(np.zeros(128)) # variable also support initialization using a numpy data + self.D_W1 = pd.Variable(shape=[784, 128], data=pd.gaussian_normal_randomizer()) + self.D_b1 = pd.Variable(np.zeros(128)) # variable also support initialization using a numpy data + self.D_W2 = pd.Varialble(np.random.rand(128, 1)) + self.D_b2 = pd.Variable(np.zeros(128)) + self.theta_D = [self.D_W0, self.D_b0, self.D_W1, self.D_b1, self.D_W2, self.D_b2] + + # define parameters of generators + self.G_W0 = pd.Variable(shape=[784, 128], data=pd.gaussian_normal_randomizer()) + self.G_b0 = pd.Variable(np.zeros(128)) # variable also support initialization using a numpy data + self.G_W1 = pd.Variable(shape=[784, 128], data=pd.gaussian_normal_randomizer()) + self.G_b1 = pd.Variable(np.zeros(128)) # variable also support initialization using a numpy data + self.G_W2 = pd.Varialble(np.random.rand(128, 1)) + self.G_b2 = pd.Variable(np.zeros(128)) + self.theta_G = [self.G_W0, self.G_b0, self.G_W1, self.G_b1, self.G_W2, self.G_b2] +``` + +### Class member Function: Generator +- Given a noisy input z, returns a fake image. +- Concatenation, batch-norm, FC operations required; +- Deconv layer required, which is missing now... +```python +class DCGAN(object): + def generator(self, z, y = None): + # input z: the random noise + # input y: input data label (optional) + # output G_im: generated fake images + + if not self.y_dim: + z = pd.layer.concat(1, [z, y]) + + G_h0 = pd.layer.fc(z, self.G_w0, self.G_b0) + G_h0_bn = pd.layer.batch_norm(G_h0) + G_h0_relu = pd.layer.relu(G_h0_bn) + + G_h1 = pd.layer.deconv(G_h0_relu, self.G_w1, self.G_b1) + G_h1_bn = pd.layer.batch_norm(G_h1) + G_h1_relu = pd.layer.relu(G_h1_bn) + + G_h2 = pd.layer.deconv(G_h1_relu, self.G_W2, self.G_b2)) + G_im = pd.layer.tanh(G_im) + return G_im +``` + +### Class member function: Discriminator +- Given a noisy input z, returns a fake image. +- Concatenation, Convolution, batch-norm, FC, Leaky-ReLU operations required; +```python +class DCGAN(object): + def discriminator(self, image): + # input image: either generated images or real ones + # output D_h2: binary logit of the label + + D_h0 = pd.layer.conv2d(image, w=self.D_w0, b=self.D_b0) + D_h0_bn = pd.layer.batchnorm(h0) + D_h0_relu = pd.layer.lrelu(h0_bn) + + D_h1 = pd.layer.conv2d(D_h0_relu, w=self.D_w1, b=self.D_b1) + D_h1_bn = pd.layer.batchnorm(D_h1) + D_h1_relu = pd.layer.lrelu(D_h1_bn) + + D_h2 = pd.layer.fc(D_h1_relu, w=self.D_w2, b=self.D_b2) + return D_h2 +``` + +### Class member function: Build the model +- Define data readers as placeholders to hold the data; +- Build generator and discriminators; +- Define two training losses for discriminator and generator, respectively. +If we have execution dependency engine to back-trace all tensors, the module building our GAN model will be like this: +```python +class DCGAN(object): + def build_model(self): + if self.y_dim: + self.y = pd.data(pd.float32, [self.batch_size, self.y_dim]) + self.images = pd.data(pd.float32, [self.batch_size, self.im_size, self.im_size]) + self.faked_images = pd.data(pd.float32, [self.batch_size, self.im_size, self.im_size]) + self.z = pd.data(tf.float32, [None, self.z_size]) + + # step 1: generate images by generator, classify real/fake images with discriminator + if self.y_dim: # if conditional GAN, includes label + self.G = self.generator(self.z, self.y) + self.D_t = self.discriminator(self.images) + # generated fake images + self.sampled = self.sampler(self.z, self.y) + self.D_f = self.discriminator(self.G) + else: # original version of GAN + self.G = self.generator(self.z) + self.D_t = self.discriminator(self.images) + # generate fake images + self.sampled = self.sampler(self.z) + self.D_f = self.discriminator(self.images) + + # step 2: define the two losses + self.d_loss_real = pd.reduce_mean(pd.cross_entropy(self.D_t, np.ones(self.batch_size)) + self.d_loss_fake = pd.reduce_mean(pd.cross_entropy(self.D_f, np.zeros(self.batch_size)) + self.d_loss = self.d_loss_real + self.d_loss_fake + + self.g_loss = pd.reduce_mean(pd.cross_entropy(self.D_f, np.ones(self.batch_szie)) +``` + +If we do not have dependency engine but blocks, the module building our GAN model will be like this: +```python +class DCGAN(object): + def build_model(self, default_block): + # input data in the default block + if self.y_dim: + self.y = pd.data(pd.float32, [self.batch_size, self.y_dim]) + self.images = pd.data(pd.float32, [self.batch_size, self.im_size, self.im_size]) + # self.faked_images = pd.data(pd.float32, [self.batch_size, self.im_size, self.im_size]) + self.z = pd.data(tf.float32, [None, self.z_size]) + + # step 1: generate images by generator, classify real/fake images with discriminator + with pd.default_block().g_block(): + if self.y_dim: # if conditional GAN, includes label + self.G = self.generator(self.z, self.y) + self.D_g = self.discriminator(self.G, self.y) + else: # original version of GAN + self.G = self.generator(self.z) + self.D_g = self.discriminator(self.G, self.y) + self.g_loss = pd.reduce_mean(pd.cross_entropy(self.D_g, np.ones(self.batch_szie)) + + with pd.default_block().d_block(): + if self.y_dim: # if conditional GAN, includes label + self.D_t = self.discriminator(self.images, self.y) + self.D_f = self.discriminator(self.G, self.y) + else: # original version of GAN + self.D_t = self.discriminator(self.images) + self.D_f = self.discriminator(self.G) + + # step 2: define the two losses + self.d_loss_real = pd.reduce_mean(pd.cross_entropy(self.D_t, np.ones(self.batch_size)) + self.d_loss_fake = pd.reduce_mean(pd.cross_entropy(self.D_f, np.zeros(self.batch_size)) + self.d_loss = self.d_loss_real + self.d_loss_fake +``` +Some small confusion and problems with this design: +- D\_g and D\_f are actually the same thing, but has to be written twice; i.e., if we want to run two sub-graphs conceptually, the same codes have to be written twice if they are shared by the graph. +- Requires ability to create a block anytime, rather than in if-else or rnn only; + +## Main function for the demo: +Generally, the user of GAN just need to the following things: +- Define an object as DCGAN class; +- Build the DCGAN model; +- Specify two optimizers for two different losses with respect to different parameters. +```python +# pd for short, should be more concise. +from paddle.v2 as pd +import numpy as np +import logging + +if __name__ == "__main__": + # dcgan class in the default graph/block + # if we use dependency engine as tensorflow + # the codes, will be slightly different like: + # dcgan = DCGAN() + # dcgan.build_model() + with pd.block() as def_block: + dcgan = DCGAN() + dcgan.build_model(def_block) + + # load mnist data + data_X, data_y = self.load_mnist() + + # Two subgraphs required!!! + with pd.block().d_block(): + d_optim = pd.train.Adam(lr = .001, beta= .1) + d_step = d_optim.minimize(dcgan.d_loss, dcgan.theta_D) + with pd.block.g_block(): + g_optim = pd.train.Adam(lr = .001, beta= .1) + g_step = pd.minimize(dcgan.g_loss, dcgan.theta_G) + + # executor + sess = pd.executor() + + # training + for epoch in xrange(10000): + for batch_id in range(N / batch_size): + idx = ... + # sample a batch + batch_im, batch_label = data_X[idx:idx+batch_size], data_y[idx:idx+batch_size] + # sample z + batch_z = np.random.uniform(-1., 1., [batch_size, z_dim]) + + if batch_id % 2 == 0: + sess.run(d_step, + feed_dict = {dcgan.images: batch_im, + dcgan.y: batch_label, + dcgan.z: batch_z}) + else: + sess.run(g_step, + feed_dict = {dcgan.z: batch_z}) +``` + +# More thinking about dependency engine v.s. block design: +- What if we just want to run an intermediate result? Do we need to run the whole block/graph? +- Should we call eval() to get the fake images in the first stage? And then train the discriminator in the second stage? diff --git a/doc/design/graph_survey.md b/doc/design/graph_survey.md new file mode 100644 index 0000000000000000000000000000000000000000..6c6db08f463ae0a2b94fc4546f123a1d7c151870 --- /dev/null +++ b/doc/design/graph_survey.md @@ -0,0 +1,232 @@ +## Survey on Graph + +Neural network framework often provides symbolic API for users to write network topology conveniently. This doc manily focus on symbolic API in most popular neural network frameworks, and try to find out how to parse symbolic configuration to a portable file, such as protobuf or json. + +### Mxnet + +The core concept of symbolic API is `Symbol`. Mxnet implements `Symbol` class in C++, and export to Python using C-API. Please refer to the comments in Mxnet: + + +`Symbol` is help class used to represent the operator node in Graph. +`Symbol` acts as an interface for building graphs from different components like Variable, Functor and Group. `Symbol` is also exported to python front-end (while Graph is not) to enable quick test and deployment. Conceptually, symbol is the final operation of a graph and thus including all the information required (the graph) to evaluate its output value. + + +A simple network topology wrote by Symbol is as follows: + +```python +def get_symbol(num_classes=10, **kwargs): + data = mx.symbol.Variable('data') + data = mx.symbol.Flatten(data=data) + fc1 = mx.symbol.FullyConnected(data = data, name='fc1', num_hidden=128) + act1 = mx.symbol.Activation(data = fc1, name='relu1', act_type="relu") + fc2 = mx.symbol.FullyConnected(data = act1, name = 'fc2', num_hidden = 64) + act2 = mx.symbol.Activation(data = fc2, name='relu2', act_type="relu") + fc3 = mx.symbol.FullyConnected(data = act2, name='fc3', num_hidden=num_classes) + mlp = mx.symbol.SoftmaxOutput(data = fc3, name = 'softmax') + return mlp +``` + + + +Varible here is actually a Symbol. Every basic Symbol will correspond to one Node, and every Node has its own NodeAttr. There is a op field in NodeAttr class, when a Symbol represents Variable(often input data), the op field is null. + +Symbol contains a data member, std::vector outputs, and NodeEntry cantains a poniter to Node. We can follow the Node pointer to get all the Graph. + +And Symbol can be saved to a Json file. + +Here is a detailed example: + +``` +>>> import mxnet as mx +>>> data = mx.symbol.Variable('data') +>>> print data.debug_str() +Variable:data + +>>> data = mx.symbol.Flatten(data=data) +>>> print data.debug_str() +Symbol Outputs: + output[0]=flatten0(0) +Variable:data +-------------------- +Op:Flatten, Name=flatten0 +Inputs: + arg[0]=data(0) version=0 + +>>> fc1 = mx.symbol.FullyConnected(data = data, name='fc1', num_hidden=128) +>>> print fc1.debug_str() +Symbol Outputs: + output[0]=fc1(0) +Variable:data +-------------------- +Op:Flatten, Name=flatten0 +Inputs: + arg[0]=data(0) version=0 +Variable:fc1_weight +Variable:fc1_bias +-------------------- +Op:FullyConnected, Name=fc1 +Inputs: + arg[0]=flatten0(0) + arg[1]=fc1_weight(0) version=0 + arg[2]=fc1_bias(0) version=0 +Attrs: + num_hidden=128 + +``` + + +### TensorFlow + + +The core concept of symbolic API is `Tensor`. Tensorflow defines `Tensor` in Python. Please refer to the comments in TensorFlow: + +A `Tensor` is a symbolic handle to one of the outputs of an `Operation`. It does not hold the values of that operation's output, but instead provides a means of computing those values in a TensorFlow [Session](https://www.tensorflow.org/api_docs/python/tf/Session). + +A simple example is as follows: + +```python + # Build a dataflow graph. + c = tf.constant([[1.0, 2.0], [3.0, 4.0]]) + d = tf.constant([[1.0, 1.0], [0.0, 1.0]]) + e = tf.matmul(c, d) + + # Construct a `Session` to execute the graph. + sess = tf.Session() + + # Execute the graph and store the value that `e` represents in `result`. + result = sess.run(e) +``` + + +The main method of `Tensor` is as follows: + + +```python +@property +def op(self): + """The `Operation` that produces this tensor as an output.""" + return self._op + +@property +def dtype(self): + """The `DType` of elements in this tensor.""" + return self._dtype + +@property +def graph(self): + """The `Graph` that contains this tensor.""" + return self._op.graph + +@property +def name(self): + """The string name of this tensor.""" + if not self._op.name: + raise ValueError("Operation was not named: %s" % self._op) + return "%s:%d" % (self._op.name, self._value_index) + +@property +def device(self): + """The name of the device on which this tensor will be produced, or None.""" + return self._op.device +``` + + +Tensor can be taken as target to run by session. Tensor contains all the information of Graph, and tracks data dependency. + + +Here is a detailed example: + + +``` +>>> import tensorflow as tf +>>> c = tf.constant([[1.0, 2.0], [3.0, 4.0]]) +>>> print c.graph + +>>> d = tf.constant([[1.0, 1.0], [0.0, 1.0]]) +>>> print d.graph + +>>> e = tf.matmul(c, d) +>>> print e.graph + +``` + +### Dynet + + +The core concept of symbolic API is `Expression`, and Dynet defines `Expression` class in C++. + + +A simple example is as follows: + +```cpp +ComputationGraph cg; +Expression W = parameter(cg, pW); + +Expression in = input(cg, xs[i]); +Expression label = input(cg, ys[i]); +Expression pred = W * in; +Expression loss = square(pred - label); +``` + +The input data and parameter are also represented by Expression. Every basci Expression corresponds to a Node. And input data is also a Node. + +Expression has a data member ComputationGraph, and ComputationGraph will be modified in users' configuring process. Expression can be a running target, beacuse Expression contains all dependency. + + +Here is a detailed example: + +write topology in C++ + +``` +ComputationGraph cg; +Expression W = parameter(cg, pW); +cg.print_graphviz(); + +Expression pred = W * xs[i]; +cg.print_graphviz(); + +Expression loss = square(pred - ys[i]); +cg.print_graphviz(); +``` + +compile and print + +``` +# first print +digraph G { + rankdir=LR; + nodesep=.05; + N0 [label="v0 = parameters({1}) @ 0x7ffe4de00110"]; +} +# second print +digraph G { + rankdir=LR; + nodesep=.05; + N0 [label="v0 = parameters({1}) @ 0x7ffe4de00110"]; + N1 [label="v1 = v0 * -0.98"]; + N0 -> N1; +} +# third print +digraph G { + rankdir=LR; + nodesep=.05; + N0 [label="v0 = parameters({1}) @ 0x7ffe4de00110"]; + N1 [label="v1 = v0 * -0.98"]; + N0 -> N1; + N2 [label="v2 = -1.88387 - v1"]; + N1 -> N2; + N3 [label="v3 = -v2"]; + N2 -> N3; + N4 [label="v4 = square(v3)"]; + N3 -> N4; +} +``` + +### Conclusion + + +Actually, Symbol/Tensor/Expression in Mxnet/TensorFlow/Dynet are the same level concepts. We use a unified name Expression here, this level concept has following features: + +- Users wirte topoloy with symbolic API, and all return value is Expression, including input data and parameter. +- Expression corresponds with a global Graph, and Expression can also be composed. +- Expression tracks all dependency and can be taken as a run target diff --git a/doc/design/if_else_op.md b/doc/design/if_else_op.md index 954a19c0733358c235eae3cffe134c23dac94c95..26d140f06db4ecefa86be015eaa731ffddc6910c 100644 --- a/doc/design/if_else_op.md +++ b/doc/design/if_else_op.md @@ -1,41 +1,51 @@ -IfOp should have only one branch. An IfOp operator takes a `cond` variable whose value must be a vector of N boolean elements. Its return value has N instances. If cond[i] == True, input instance input[i] will go through true_block() and generate output[i]; otherwise it will produce output from false_bloack(). +# The `IfElse` Operator -```python -import paddle as pd +PaddlePaddle's `IfElse` operator differs from TensorFlow's: -x = var() -y = var() -cond = var() -default_value = var() -b = pd.create_ifelseop(inputs=[x], output_num=1) -with b.true_block(): - x = b.inputs(0) - z = operator.add(x, y) - b.set_output(0, operator.softmax(z)) - -with b.false_block(): - x = b.inputs(0) - z = layer.fc(x) - b.set_output(0, operator.softmax(z)) - -out = b(cond) -``` +- the TensorFlow version takes a scalar boolean value as the condition so that the whole mini-batch goes to either the true or the false branch, whereas +- the PaddlePaddle version takes a vector of boolean value as the condition, and instances corresponding to true values go to the true branch, those corresponding to false values go to the false branch. + +## Example + +The following PaddlePaddle program shows the usage of the IfElse operator: -If only true_block is set in an IfElseOp, a special case is that we can have a default value for false as: ```python import paddle as pd -x = var() -y = var() -cond = var() -default_value = var() -b = pd.create_ifelseop(inputs=[x], output_num=1, default_value) - -with b.true_block(): - x = b.inputs(0) - z = operator.add(x, y) - b.set_output(0, operator.softmax(z)) +x = minibatch([10, 20, 30]) # shape=[None, 1] +y = var(1) # shape=[1], value=1 +z = minibatch([10, 20, 30]) # shape=[None, 1] +cond = larger_than(x, 15) # [false, true, true] + +ie = pd.ifelse() +with ie.true_block(): + d = pd.layer.add(x, y) + ie.output(d, pd.layer.softmax(d)) +with ie.false_block(): + d = pd.layer.fc(z) + ie.output(d, d+1) +o1, o2 = ie(cond) +``` -out = b(cond) +A challenge to implement the `IfElse` operator is to infer those variables to be split, or, say, to identify the variable of the mini-batch or those derived from the mini-batch. + +An equivalent C++ program is as follows: + +```c++ +namespace pd = paddle; + +int x = 10; +int y = 1; +int z = 10; +bool cond = false; +int o1, o2; +if (cond) { + int d = x + y; + o1 = z; + o2 = pd::layer::softmax(z); +} else { + int d = pd::layer::fc(z); + o1 = d; + o2 = d+1; +} ``` -where default_value is a list of vars for `cond` == False. diff --git a/doc/design/images/asgd.gif b/doc/design/images/asgd.gif new file mode 100644 index 0000000000000000000000000000000000000000..4a0da7bf6df9326a2aab1638b77c5455c18b8c4e Binary files /dev/null and b/doc/design/images/asgd.gif differ diff --git a/doc/design/images/feed_forward.png b/doc/design/images/feed_forward.png new file mode 100644 index 0000000000000000000000000000000000000000..d312371a04c26aa6cd196e0bd1f51becb425180b Binary files /dev/null and b/doc/design/images/feed_forward.png differ diff --git a/doc/design/images/feed_forward_regularized.png b/doc/design/images/feed_forward_regularized.png new file mode 100644 index 0000000000000000000000000000000000000000..677e99bfd9f8e72ed9fe4b27127af2ced202f447 Binary files /dev/null and b/doc/design/images/feed_forward_regularized.png differ diff --git a/doc/design/images/graph_construction_example.dot b/doc/design/images/graph_construction_example.dot index 8d1b673abf6b78c851676fa379dc850c4818f0e5..e115f9844bae6ad24f638c8ed4749cea8aff06a9 100644 --- a/doc/design/images/graph_construction_example.dot +++ b/doc/design/images/graph_construction_example.dot @@ -33,7 +33,6 @@ digraph ImageClassificationGraph { cost -> MSE_Grad [color=red]; d_cost -> MSE_Grad [color=red]; - x -> MSE_Grad [color=red]; l -> MSE_Grad [color=red]; y -> MSE_Grad -> d_y [color=red]; diff --git a/doc/design/images/graph_construction_example_all.png b/doc/design/images/graph_construction_example_all.png index 181187503472d15779b87284105841168b3945c4..261611a5721f9aa97874f7e6d897fe48cf667db2 100644 Binary files a/doc/design/images/graph_construction_example_all.png and b/doc/design/images/graph_construction_example_all.png differ diff --git a/doc/design/images/graph_construction_example_forward_backward.png b/doc/design/images/graph_construction_example_forward_backward.png index 3049a9315fd616464dec54e33064cb75598ca536..4c69687f4a6a181138f3df72ce5e8aa48487b5be 100644 Binary files a/doc/design/images/graph_construction_example_forward_backward.png and b/doc/design/images/graph_construction_example_forward_backward.png differ diff --git a/doc/design/images/graph_construction_example_forward_only.png b/doc/design/images/graph_construction_example_forward_only.png index 25d19088cbf0b5f68cf734f2ff21eba8af4a2860..e668c16e0cac73acb4e5dc2b1827557ae77126b4 100644 Binary files a/doc/design/images/graph_construction_example_forward_only.png and b/doc/design/images/graph_construction_example_forward_only.png differ diff --git a/doc/design/images/l1_regularization.png b/doc/design/images/l1_regularization.png new file mode 100644 index 0000000000000000000000000000000000000000..e1b9c7a44f94dc027598a98da93ddb8133190972 Binary files /dev/null and b/doc/design/images/l1_regularization.png differ diff --git a/doc/design/images/l2_regularization.png b/doc/design/images/l2_regularization.png new file mode 100644 index 0000000000000000000000000000000000000000..d5c2fcbc2ccae75ad083162e5a2dceb0210be298 Binary files /dev/null and b/doc/design/images/l2_regularization.png differ diff --git a/doc/design/images/loss_equation.png b/doc/design/images/loss_equation.png new file mode 100644 index 0000000000000000000000000000000000000000..14212ec8d36c803de96bde8a9a4b5591bd20434e Binary files /dev/null and b/doc/design/images/loss_equation.png differ diff --git a/doc/design/images/theta_star.gif b/doc/design/images/theta_star.gif new file mode 100644 index 0000000000000000000000000000000000000000..dd24d33e124396be3fc410c9b12f33148f64efe2 Binary files /dev/null and b/doc/design/images/theta_star.gif differ diff --git a/doc/design/infer_var_type.md b/doc/design/infer_var_type.md new file mode 100644 index 0000000000000000000000000000000000000000..d9d5397becba2ef1806d9341cd49cd9aabbf4a6a --- /dev/null +++ b/doc/design/infer_var_type.md @@ -0,0 +1,78 @@ +# Design Doc: InferVarType + +## The Problem Posed + +The variable in our design can hold variant types. Such as `LoDTensor` and `SelectedRows`. An operator should be able to inference the variable types of its output. + +For example, a `lookup table` operator takes two `LoDTensor`; one is a float tensor as the embedding table, the other is an int tensor as word ID. The gradient operator of `lookup table` will generate a `SelectedRows` as its output. A `sum` operator can take both `LoDTensor` and `SelectedRows` as its inputs and will generate a `LoDTensor` if any of its inputs is `LoDTensor`, otherwise, the `sum` operator will generate `SelectedRows` as its output. + +The variable type will be constant at runtime. Every variable's type can either be set by the user (input data and parameter) or be inferred by the operator in compile time. + +## Proposed Solution + +The `InferVarType` is a compile-time function which is registered to each operator. The inferface of that function is: + + +```c++ +using InferVarTypeFN = std::function< + void (const OpDescBind& /*op_desc*/, BlockDescBind* /*block*/)>; +``` + +It takes an operator description as its input and will write the output variable type and store them in block description. + +The `InferVarTypeFN` will be registered in `OpInfo`, to replace `infer_var_type_` field. The `OpInfo` should be + +```cpp +struct OpInfo { + InferVarTypeFN infer_var_type_; + ... +}; +``` + +The default `InferVarType` will set output type as `LoDTensor`. It can be done by `GetInferVarType()`. + +```cpp +void DefaultInferVarType(const OpDescBind& op_desc, BlockDescBind* block) { + // set the output type of variable as `LoDTensor`. + // ... +} + +struct OpInfo { + InferVarTypeFN infer_var_type_; + InferVarTypeFN GetInferVarType() const { + if (infer_var_type_) { + return infer_var_type_; + } else { + return DefaultInferVarType; + } + } +}; +``` + +## Register InferVarType + +We provide a thin base class for registering an `InferVarTypeFN`. To use a base class will ease the implementation of registry since we can detect the registry entry is an `InferVarTypeFN` or not. + +```cpp +class VarTypeInferer { +public: + virtual void operator()(const OpDescBind& op_desc, BlockDescBind* block) const = 0; +} +``` + +Operator developers can write the specialize `VarTypeInferer` as follow. + +```cpp +class SpecialVarTypeInferer : public VarTypeInferer { +public: + virtual void operator()(const OpDescBind& op_desc, BlockDescBind* block) const { + // .. own logic + } +} +``` + +Then user can register the `InferVarType` just like `GradOpDescMaker` and `OpInfoMaker`. + +``` +REGISTER_OPERATOR(some_op, OpType, SpecialVarTypeInferer, ...); +``` diff --git a/doc/design/model_format.md b/doc/design/model_format.md new file mode 100644 index 0000000000000000000000000000000000000000..e29129fddf775939c9f7a8b49d850d523e6e5a45 --- /dev/null +++ b/doc/design/model_format.md @@ -0,0 +1,36 @@ +# Design Doc: Model Format + +## Motivation + +A model is an output of the training process. One complete model consists of two parts, the **topology** and the **parameters**. In order to support industrial deployment, the model format must be self-complete and must not expose any training source code. + +As a result, In PaddlePaddle, the **topology** is represented as a [ProgramDesc](https://github.com/PaddlePaddle/Paddle/blob/1c0a4c901c9fc881d120249c703b15d1c50dae7d/doc/design/program.md), which describes the model structure. The **parameters** contain all the trainable weights in the model. We must support large size parameters and efficient serialization/deserialization of parameters. + +## Implementation + +The topology is saved as a plain text in a detailed self-contain protobuf file. + +The parameters are saved as a binary file. As we all know, the protobuf message has a limit of [64M size](https://developers.google.com/protocol-buffers/docs/reference/cpp/google.protobuf.io.coded_stream#CodedInputStream.SetTotalBytesLimit.details). We have done a [benchmark experiment](https://github.com/PaddlePaddle/Paddle/pull/4610), which shows that protobuf is not fit for the task. + +As a result, we design a particular format for tensor serialization. By default, an arbitrary tensor in Paddle is a [LoDTensor](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/framework/lod_tensor.md), and has a description information proto of [LoDTensorDesc](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/framework/framework.proto#L99). We save the DescProto as the byte string header. It contains all the necessary information, such as the `dims`, and the `LoD` information in [LoDTensor](https://github.com/PaddlePaddle/Paddle/blob/1c0a4c901c9fc881d120249c703b15d1c50dae7d/paddle/framework/lod_tensor.md). A tensor stores values in a continuous memory buffer. For speed we dump the raw memory to disk and save it as the byte string content. So, the binary format of one tensor is, + +The table below shows a tensor's byte view in detail. Note that all the signed values are written in the little-endian format. + +|field name | type | description | +| --- | --- | --- | +| version | uint32_t | Version of saved file. Always 0 now. | +| tensor desc length | uint32_t | TensorDesc(Protobuf message) length in bytes. | +| tensor desc | void* | TensorDesc protobuf binary message | +| tensor data | void* | Tensor's data in binary format. The length of `tensor_data` is decided by `TensorDesc.dims()` and `TensorDesc.data_type()` | +| lod_level | uint64_t | Level of LoD | +| length of lod[0] | uint64_t | [Optional] length of lod[0] in bytes. | +| data of lod[0] | uint64_t* | [Optional] lod[0].data() | +| ... | ... | ... | + + + +## Summary + +- We introduce a model format. +- The model represented by its forward-pass computation procedure is saved in a **ProgramDesc** protobuf message. +- A bunch of specified format binary tensors describe the **parameters**. diff --git a/doc/design/optimizer.md b/doc/design/optimizer.md new file mode 100644 index 0000000000000000000000000000000000000000..202b4b65103c0b7c536a9cb466c4120ce134d8c3 --- /dev/null +++ b/doc/design/optimizer.md @@ -0,0 +1,91 @@ +## Optimizer Design + +### The Problem + +A PaddlePaddle program, or a block, is a sequence of operators operating variables. A training program needs to do three kinds of works: + +1. the forward pass, which computes intermediate results and the cost(s), +1. the backward pass, which derives gradients from intermediate results and costs, and +1. the optimization pass, which update model parameters to optimize the cost(s). + +These works rely on three kinds of operators: + +1. forward operators, +1. gradient operators, and +1. optimization operators. + +It's true that users should be able to create all these operators manually by calling some low-level API, but it would be much more convenient if they could only describe the forward pass and let PaddlePaddle create the backward and optimization operators automatically. + +In this design, we propose a high-level API that automatically derives the optimisation pass and operators from the forward pass. + + +### High-level Python API to describe the training process + +1. User write code to describe the network: + + ```python + images = layer.data("images") + labels = layer.data("labels") + w1 = pd.var("w1") + b1 = pd.var("b1") + hidden = layer.fc(images, w=w1, b=b1) + cost = layer.mse(hidden, labels) + ``` + + The above code snippet will create forward operators in [Block](https://github.com/PaddlePaddle/Paddle/blob/develop/doc/design/block.md). + + +2. Users create a certain kind of Optimizer with some argument. + + ```python + optimizer = AdagradOptimizer(learing_rate=0.001) + ``` + +3. Users use the optimizer to `minimize` a certain `cost` through updating parameters in parameter_list. + + ```python + opt_op_list = optimizer.minimize(cost, parameter_list=[w1, b1]) + ``` + The above code snippet will create gradient and optimization operators in Block. The return value of `minimize()` is list of optimization operators that will be run by session. + +4. Users use Session/Executor to run this opt_op_list as target to do training. + + ```python + sess.run(target= opt_op_list, ...) + ``` + +#### Optimizer Python interface: + +```python +class Optimizer(object): + """Optimizer Base class. + + """ + + def __init__(self): + pass + + def create_optimization_pass(self, parameters_and_grads): + """Add optimization operators to update gradients to variables. + + Args: + parameters_and_grads: a list of (variable, gradient) pair to update. + + Returns: + optmization_op_list: a list of optimization operator that will update parameter using gradient. + """ + return None + + def minimize(self, loss, parameter_list): + """Add operations to minimize `loss` by updating `parameter_list`. + + This method combines interface `append_backward_ops()` and + `create_optimization_pass()` into one. + """ + params_grads = self.create_backward_pass(loss, parameter_list) + update_ops = self.create_optimization_pass(params_grads) + return update_ops + +``` + +Users can inherit the Optimizer above to create their own Optimizer with some special logic, such as AdagradOptimizer. diff --git a/doc/design/parameter_average.md b/doc/design/parameter_average.md new file mode 100644 index 0000000000000000000000000000000000000000..2c4edee9fe31d502ea62b9fe5c8757c0a4c5e79f --- /dev/null +++ b/doc/design/parameter_average.md @@ -0,0 +1,72 @@ +# Averaging Parameter in PaddlePaddle + +## Why Averaging +In a large scale machine learning setup where the size of the training data is huge, it could take us a large number of iterations over the training data before we can achieve the optimal values of parameters of our model. Looking at the problem setup, it is desirable if we can obtain the optimal values of parameters by going through the data in as few passes as we can. + +Polyak and Juditsky (1992) showed that the test performance of simple average of parameters obtained by Stochastic Gradient Descent (SGD) is as good as that of parameter values that are obtained by training the model over and over again, over the training dataset. + +Hence, to accelerate the speed of Stochastic Gradient Descent, Averaged Stochastic Gradient Descent (ASGD) was proposed in Polyak and Juditsky (1992). For ASGD, the running average of parameters obtained by SGD, is used as the estimator for
. The averaging is done as follows: + +
+ +We propose averaging for any optimizer similar to how ASGD performs it, as mentioned above. + +### How to perform Parameter Averaging in PaddlePaddle + +Parameter Averaging in PaddlePaddle works in the following way during training : +1. It will take in an instance of a normal optimizer as an input, e.g. RMSPropOptimizer +2. The optimizer itself is responsible for updating the parameters. +3. The ParameterAverageOptimizer maintains a separate copy of the parameters for itself: + 1. In concept, the values of this copy are the average of the values of the parameters in the most recent N batches. + 2. However, saving all the N instances of the parameters in memory is not feasible. + 3. Therefore, an approximation algorithm is used. + +Hence, overall we have have two copies of the parameters: one for the optimizer itself, and one for the ParameterAverageOptimizer. The former should be used in back propagation, while the latter should be used during testing and should be saved. + +During the testing/ saving the model phase, we perform the following steps: +1. Perform the delayed operations. +2. Save current values of the parameters to a temporary variable. +3. Replace the values of the parameters with the averaged values. +4. Perform testing and/or save the parameters. +5. Restore the values of the parameters once done. + +### How to implement Averaging of Parameter in PaddlePaddle + +We can add the ParameterAverageOptimizer op to the graph through Python API. Using this approach, we manually add this op to the graph and direct the output of the optimizer op to this op during training. + + **Advantages**: + - Allows for greater flexibility to the users of PaddlePaddle. Using this approach, the users can plug different optimizers into ParameterAverageOptimizer by passing in the optimizer to the op. + - Makes it easy for the users to customize and extend the framework. + + **Disadvantages**: + - Implementation requires re-writing the averaging methodology in Python. + +### Low-Level implementation + +In the new design, we propose to create a new operation for averaging parameter updates (ParameterAverageOptimizer). For now, we can add an op that takes in the following as input: +- the optimizer +- the window_size to keep the updates + +The ParameterAverageOptimizer op can be like any other operator with its own CPU/GPU implementation either using Eigen or separate CPU and GPU kernels. As the initial implementation, we can implement the kernel using Eigen following the abstraction pattern implemented for [Operators](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/operators/rmsprop_op.h). We also want to support the case when the Trainer/Optimizer runs on the GPU while ParameterAverageOptimizer runs on a CPU. + +The idea of building an op for averaging is in sync with the refactored PaddlePaddle philosophy of using operators to represent any computation unit. The way the op will be added to the computation graph will be decided by the [layer functions](https://github.com/PaddlePaddle/Paddle/blob/develop/doc/design/python_api.md#layer-function) in Python API. + +### Python API implementation for ParameterAverageOptimizer + +Based on Polyak and Juditsky (1992), we can generalize the averaging of updates to any optimizer. The input to the op would be the following: +- Any optimizer (RMSProp , AdaGrad etc.) +- A window size. The op keeps accumulating updated parameter values over a window of N batches and takes an average. Move the averaged value to a buffer when window is full to avoid loss of precision. + +Using the ParameterAverageOptimizer op, any user can add the operation to their computation graphs. However, this will require a lot of lines of code and we should design Python APIs that support averaging. As per the PaddlePaddle [Python API design](https://github.com/PaddlePaddle/Paddle/blob/develop/doc/design/python_api.md), the layer functions are responsible for creating operators, operator parameters and variables. Since ParameterAverageOptimizer will be an operator, it makes sense to create it in the layer functions. +We will have a wrapper written in Python that will support the functionality and implement the actual core computation in C++ core as we have done for other [Optimizers](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/operators/rmsprop_op.cc) + +#### Creation of the ParameterAverageOptimizer operator +There are two ways for creating the ParameterAverageOptimizer op: +1. We create the op immediately while building the computation graph. +2. We add the op in a lazy manner, just before the backward pass, similar to the way the optimization ops are added. + +The proposal is to add the op immediately while building the computation graph. + +#### High-level API + +In PaddlePaddle Python API, users will primarily rely on [layer functions](https://github.com/PaddlePaddle/Paddle/blob/develop/doc/design/python_api.md#layer-function) to create neural network layers. Hence, we also need to provide parameter average functionality in layer functions. diff --git a/doc/design/program.md b/doc/design/program.md new file mode 100644 index 0000000000000000000000000000000000000000..bd2456787c4e336d357a65255a8274a7c9e465cc --- /dev/null +++ b/doc/design/program.md @@ -0,0 +1,139 @@ +# Design Doc: PaddlePaddle Programs + +## Compile and Execution + +A PaddlePaddle program consists of two parts -- the first generates a `ProgramDesc` protobuf message that describes the program, and the second runs this message using a C++ class `Executor`. + +A simple example PaddlePaddle program can be found in [graph.md](./graph.md): + +```python +x = layer.data("images") +l = layer.data("label") +y = layer.fc(x) +cost = layer.mse(y, l) +optimize(cost) +train(cost, reader=mnist.train()) +``` + +The first five lines of the following PaddlePaddle program generates, or, compiles, the `ProgramDesc` message. The last line runs it. + +## Programs and Blocks + +The basic structure of a PaddlePaddle program is some nested blocks, as a C++ or Java program. + +- program: some nested blocks +- [block](./block.md): + - some local variable definitions, and + - a sequence of operators + +The concept of block comes from usual programs. For example, the following C++ program has three blocks: + +```c++ +int main() { // block 0 + int i = 0; + if (i < 10) { // block 1 + for (int j = 0; j < 10; j++) { // block 2 + } + } + return 0; +} +``` + +The following PaddlePaddle program has three blocks: + +```python +import paddle as pd // block 0 + +x = minibatch([10, 20, 30]) # shape=[None, 1] +y = var(1) # shape=[1], value=1 +z = minibatch([10, 20, 30]) # shape=[None, 1] +cond = larger_than(x, 15) # [false, true, true] + +ie = pd.ifelse() +with ie.true_block(): // block 1 + d = pd.layer.add_scalar(x, y) + ie.output(d, pd.layer.softmax(d)) +with ie.false_block(): // block 2 + d = pd.layer.fc(z) + ie.output(d, d+1) +o1, o2 = ie(cond) +``` + +## `BlockDesc` and `ProgramDesc` + +All protobuf messages are defined in `framework.proto`. + +`BlockDesc` is straight-forward -- it includes local variable definitions, `vars`, and a sequence of operators, `ops`. + +```protobuf +message BlockDesc { + required int32 parent = 1; + repeated VarDesc vars = 2; + repeated OpDesc ops = 3; +} +``` + +The parent ID indicates the parent block so that operators in a block can refer to variables defined locally and also those defined in their ancestor blocks. + +All hierarchical blocks in a program are flattened and stored in an array. The block ID is the index of the block in this array. + +```protobuf +message ProgramDesc { + repeated BlockDesc blocks = 1; +} +``` + + +### Global Block + +The global block is the first one in the above array. + +## Operators that Use Blocks + +In the above example, the operator `IfElseOp` has two blocks -- the true branch and the false branch. + +The definition of `OpDesc` shows that an operator could have some attributes: + +```protobuf +message OpDesc { + AttrDesc attrs = 1; + ... +} +``` + +and an attribute could be of type block, which is, in fact, a block ID as described above: + +``` +message AttrDesc { + required string name = 1; + + enum AttrType { + INT = 1, + STRING = 2, + ... + BLOCK = ... + } + required AttrType type = 2; + + optional int32 block = 10; // when type == BLOCK + ... +} +``` + +## InferShape + +With this design, the InferShape function should take the following parameters: + +```c++ +void InferShape(int current_block, + int current_operator, + ProgramDesc* program // might change VarDesc values. + ) { + ... +} +``` + +where + +- `current_block` indices into `ProgramDesc::blocks`, +- `current_operator` indices into `BlockDesc::ops`. diff --git a/doc/design/prune.md b/doc/design/prune.md new file mode 100644 index 0000000000000000000000000000000000000000..4a5cf10c79a554779137f0cce5494fdd96ef6b7a --- /dev/null +++ b/doc/design/prune.md @@ -0,0 +1,63 @@ +# Prune + +## Motivation + +We want to support running inference, training and checkpointing in one `ProgramDesc`. We implement +`void Prune(const ProgramDesc* input, ProgramDesc* output)` function, which takes a `ProgramDesc` +and generate a pruned `ProgramDesc`. + +## Challenge + +Pruning need to support both variables and operators being evaluation targets. Consider the following +different situations. + +```python +# Case 1: run foward pass. +cost_np = session.run(target=cost) +# Case 2: run backward passing. +opts_np, _ = session.run(target=[cost, opt]) +# Case 3: run checkpointing +_ = session.run(target=checkpoint) +``` + +## Solution + +To support evaluation of operators, we add `is_target` field in the `OpDesc`. + +```c++ +message OpDesc { + required string type = 3; + repeated Var inputs = 1; + repeated Var outputs = 2; + repeated Attr attrs = 4; + optional bool is_target = 5 [ default = false ]; +}; +``` + +To support evaluation of variables, we add [fetch_op](https://github.com/PaddlePaddle/Paddle/pull/4599). +For each variable in the `target`, we insert a `fetch_op` into the `ProgramDesc` with `variable` being +`fetch_op`'s input. Then we also set `fetch_op` is a target. + +### Algorithm + +If an operator needs to be run, it must fall into one of the following cases: + +1. It is the target. +2. It is depended by some other ops, meaning its output is some other op's input. + +The first case can be checked by `op_desc.is_traget()` . The second case can be implement as + +```c++ +bool HasDependentVar(const OpDesc& op_desc, const std::set& dependent_vars) { + for (auto& var : op_desc.outputs()) { + for (auto& argu : var.arguments()) { + if (dependent_vars.count(argu) != 0) { + return true; + } + } + } + return false; +} +``` + +Then the whole algorithm can be implemented as the following [code](https://github.com/tonyyang-svail/Paddle/blob/prune_impl/paddle/framework/prune.cc). diff --git a/doc/design/python_api.md b/doc/design/python_api.md new file mode 100644 index 0000000000000000000000000000000000000000..cb5fdc765b7126fc66a1c8978d4b96c0dc5a9f2c --- /dev/null +++ b/doc/design/python_api.md @@ -0,0 +1,284 @@ +# Design Doc: Python API + +Due to the refactorization of the PaddlePaddle core, we need Python classes to construct corresponding protobuf messages that describe a DL program. + +| Python classes | Protobuf messages | +| --- | --- | +| Program | ProgramDesc | +| Block | BlockDesc | +| Operator | OpDesc | +| Variable | VarDesc | + +Please be aware that these Python classes need to maintain some construction-time information, which are not part of the protobuf messages. + +## Core Concepts + +### Program + +A `ProgramDesc` describes a [DL program](https://github.com/PaddlePaddle/Paddle/blob/develop/doc/design/program.md), which is composed of an array of `BlockDesc`s. The `BlockDesc`s in a `ProgramDesc` can have a tree-like hierarchical structure. However, the `ProgramDesc` onlys stores a flattened array of `BlockDesc`s. A `BlockDesc` refers to its parent block by its index in the array. For example, operators in the step block of an RNN operator need to be able to access variables in its ancestor blocks. + +Whenever we create a block, we need to set its parent block to the current block, hence the Python class `Program` needs to maintain a data member `current_block`. + +```python +class Program(objects): + def __init__(self): + self.desc = core.NewProgram() # a C++ ProgramDesc pointer. + self.blocks = vector() + self.blocks.append(Block(self, -1)) # the global block + self.current_block = 0 # initialized to the global block + + def global_block(): + return self.blocks[0] + + def current_block(): + return self.get_block(self.current_block) + + def rollback(): + self.current_block = self.current_block().parent_idx + + def create_block(): + new_block_idx = len(self.block) + self.blocks.append(Block(self, self.current_block)) + self.current_block = new_block_idx + return current_block() +``` + +`Program` is an accessor to the protobuf message `ProgramDesc`, which is created in C++ space, because the InferShape function is in C++, which manipulates `VarDesc` messages, which are in turn members of `BlockDesc`, which is a member of `ProgramDesc`. + +`Program` creates the first block as the global block in its constructor. All parameters and their initializer operators are in the global block. + +### Block + +A [Block](https://github.com/PaddlePaddle/Paddle/blob/develop/doc/design/block.md) includes + +1. a map from variable names to an instance of the Python `Variable` class, and +1. a list of `Operator` instances. + +```python +class Block(objects): + def __init__(self, program, parent_idx): + self.desc = core.NewBlock(program.desc) + self.program = program + self.vars = map() + self.ops = vector() + self.parent_idx = parent_idx + + def create_var(self, ...): + return Variable(self, ...) + + def _create_global_var(self, ...): + program.global_block().create_var(...) + + def create_parameter(self, name, ...): + # Parameter is a subclass of variable. See Parameter section for details. + self.vars[name] = Parameter(self._create_global_var(...), ...) + return self.vars[name] + + def append_operator(self, ...): + self.ops.append(Operator(self, ...)) + + def prepend_operator(self, ...): # Parameter's ctor prepands initialize operators. + self.ops.prepend(Operator(self, ...)) +``` + +`create_parameter` is necessary because parameters are global variables, defined in the global block, but can be created in some sub-blocks. For example, an FC layer in the step block of an RNN operator. + +`prepend_operator` is necessary because the constructor of `Parameter` needs to create the initialize (or load) operator of the parameter, and would like to put it in the *preamble* of the global block. + +### Operator + +The `Operator` class fills in the `OpDesc` message and calls the C++ function `InferShape` to infer the output shapes from the input shapes. + +```python +class Operator(object): + def __init__(self, + block, # Block + type, # string + inputs, # dict + outputs,# dict + attrs # dict + ): + self.desc = core.NewOpDesc(block.desc, type, inputs, outputs, attrs) + core.infer_shape(self.desc, inputs, outputs) + + def type(self): + return self.desc.type() +``` + +`Operator` creates the `OpDesc` message in C++ space, so that it can call the `InferShape` function, which is in C++. + +### Variable + +Operators take Variables as its inputs and outputs. + +```python +class Variable(object): + def __init__(self, + block=None, # Block + name=None, # string + shape, # tuple + dtype="float32", # string + lod_level=None # int + ): + if name is None: + name = unique_name_generator() + self.name = name + self.block = block + self.desc = core.NewVarDesc(block.desc, name, shape, lod_level) + self.writer = None +``` + +Please be aware of `self.writer`, that tracks operator who creates the variable. It possible that there are more than one operators who write a variable, but in Python space, each write to a variable is represented by a Variable class. This is guaranteed by the fact that **`core.NewVarDesc` must NOT create a new `VarDesc` message if its name already exists in the specified block**. + +### Parameter + +A parameter is a global variable with an initializer (or load) operator. + +```python +class Parameter(Variable): + def __init__(self, + block=None, # Block + name=None, # string + shape, # tuple + dtype="float32", # string + lod_level=None # int + trainable, # bool + initialize_op_attrs, + optimize_op_attrs): + super(Parameter, self).__init__(block, name, shape, dtype, lod_level) + self.trainable = trainable + self.optimize_op_attrs = optimize_op_attrs + block.prepend(Operator(block, # Block + initialize_op_attrs['type'], # string + None, # no inputs + self, # output is the parameter + initialize_op_attrs) +``` + +When users create a parameter, they can call + +```python +program.create_parameter( + ..., + init_attr={ + type: "uniform_random", + min: -1.0, + max: 1.0, + }) +) +``` + +In above example, `init_attr.type` names an initialize operator. It can also name the load operator + +```python +init_attr={ + type: "load", + filename: "something.numpy", +} +``` + +`optimize_op_attrs` is not in the `VarDesc` message, but kept in the Python instance, as it will be used in the Python space when creating the optimize operator's `OpDesc`, and will be in the `OpDesc` message. + +## Layer Function + +A layer is a Python function that creates some operators and variables. Layers simplify the work of application programmers. + +Layer functions take `Variable` and configuration parameters as its input and return the output variable(s). + +For example, `FullyConnected` take one or more variable as its input. The input could be input data or another layer's output. There are many configuration options for a `FullyConnected` layer, such as layer size, activation, parameter names, initialization strategies of parameters, and so on. The `FullyConnected` layer will return an output variable. + + +### Necessity for reusing code between layer functions + +There are a lot of code that can be reused. Such as + +* Give the default value of configuration. e.g., default initialize strategy for parameters is uniform random with `min = -1.0`, `max = 1.0`. and default initialize strategy for bias is to fill zero. +* Append the activation operator. +* Create a temporary variable. +* Create parameter. +* Generate a unique name. +* Add a bias. +* ... + +A mechanism to reuse code between layer functions is necessary. It will be around [150 lines of code](https://github.com/PaddlePaddle/Paddle/pull/4724/files#diff-823b27e07e93914ada859232ae23f846R12) if we write a `FullyConnected` layer without any helper functions. + + + +### Comparision between global functions and helper class + +The `FullyConnected` layer will be as follow when we provide global functions: + +```python +def fc_layer(input, size, param_attr=None, bias_attr=None, act=None, name=None): + if name is None: + name = unique_name("fc") + input = multiple_input(input) + param_attr = default_param_attr(param_attr) + param_attr = multiple_param_attr(param_attr, len(input)) + + # mul + mul_results = [] + for ipt, attr in zip(input, param_attr): + shape = ipt.shape[1:] + [size] + w = g_program.global_block().create_parameter(shape, ipt.dtype, name, attr) + tmp = create_tmp_var(name) + g_program.current_block().append_op("mul", {ipt, w}, {tmp}) + mul_results.append(tmp) + + # add sum + ... + # add bias + ... + # add activation + ... + return out +``` + +We can provide many helpers functions for layer developers. However, there are several disadvantages for global helper functions: + +1. We need a namespace for these methods, then layer developers can quickly figure out what method they can use. +2. Global functions will force layer developers to pass its parameter time by time. + +So we provide a helper class, `LayerHelper`, to share code between layer functions. The `FullyConnected` Layer will be as follow. + +```python +def fc_layer(input, size, param_attr=None, bias_attr=None, act=None, name=None): + helper = LayerHelper(locals()) # pass all parameter to LayerHelper + + mul_results = [] + for ipt, param in helper.iter_multiple_input_and_param(): + w = helper.create_parameter(shape=ipt.shape[1:] + [size], dtype = ipt.dtype) + tmp = helper.create_tmp_variable() + helper.append_op('mul', {ipt, w}, {tmp}) + mul_results.append(tmp) + + pre_bias = helper.add_sum(mul_results) + pre_activation = helper.add_bias(pre_bias) + return helper.add_activation(pre_activation) +``` + +We not only use the fewer lines of code to write `fc_layer` but also make the code clearer to understand. At the same time, layer developers can figure out what function they can invoke by typing `helper.` in a python editor. + + +### Implementation of layer helper + +We just keep all parameters of a layer function as a dictionary in layer helper as a private data member. Every method of layer helper will look up the dictionary after it is invoked. In that way, we can implement a layer helper for all layer functions even some layer does not contain some operator. For example, The `activation` is used by the FullyConnected layer or convolution layers, but a cross-entropy layer does not use it. The example code of `add_activation` are: + +```python +class LayerHelper(object): + def __init__(self, **kwargs): # kwargs is short for `keyword arguments` + self.kwargs = kwargs + + def add_activation(self, input_var): + act = self.kwargs.get("act", None) # default value is None + if act is None: # do nothing if no act + return input_var + + tmp = self.create_tmp_var(self) + self.append_op(type=act, input=input_var, output=tmp) + return tmp +``` + +## Optimizer + +[Optimizer Design Doc](./optimizer.md) diff --git a/doc/design/refactor/distributed_architecture.md b/doc/design/refactor/distributed_architecture.md new file mode 100644 index 0000000000000000000000000000000000000000..ac7e98ccf1aadbb973a4801fde842375cf63448c --- /dev/null +++ b/doc/design/refactor/distributed_architecture.md @@ -0,0 +1,222 @@ +# Design Doc: Distributed Training Architecture + +## Abstract + +PaddlePaddle v0.10.0 uses the "trainer-parameter server" +architecture. We run multiple replicated instances of trainers (runs +the same code written by the user) and parameter servers for +distributed training. This architecture served us well, but has some +limitations: + +1. Need to write special code to handle tasks which should only be run + by a single trainer. E.g., initializing model and saving model. + +2. Model parallelism is hard: need to write if-else branches conditioned + on the trainer ID to partition model onto each trainer, and manually + write the inter-model-shard communication code. + +3. The user can not directly specify the parameter update rule: need + to modify the parameter server C++ code and compile a new + binary. This adds complication for researchers: A lot of extra + effort is required. Besides, the training job submission program + may not allow running arbitrary binaries. + +This design doc discusses PaddlePaddle's new distributed training +architecture that addresses the above limitations. + +## Analysis + +We will assume the user writes the trainer program by Python, the same +analysis holds if the trainer program is written in C++. + +### Limitation 1 + +If we look at the Python code that the user writes, there are two +kinds of functionalities: + +- The training logic such as load / save model and print log. +- The neural network definition such as the definition of the data + layer, the fully connected layer, the cost function and the + optimizer. + +When we training with PaddlePaddle v0.10.0 distributedly, multiple +replicated Python instances are running on different nodes: both the +training logic and the neural network computation is replicated. + +The tasks that should only run once all belong to the training logic, +if we only replicate the neural network computation, but do **not** +replicate the training logic, the limitation could be solved. + +### Limitation 2 + +Model parallelism means running a single model on multiple nodes by +partitioning the model onto different nodes and managing the +inter-model-shard communications. + +PaddlePaddle should be able to modify the nerual network computation +definition to support model parallelism automatically. However, the +computation is only specified in Python code, and PaddlePaddle can not +modify Python code. + +Just like compiler uses a intermediate representation (IR) so that +programmer does not need to manually optimize their code in most of +the cases - the compiler will optimize the IR: + + + +We can have our own IR too: PaddlePaddle can support model parallel by +converting the IR so the user no longer need to manually do it in +Python: + + + +The IR for PaddlePaddle after refactor is called `Block`, it specifies +the computation dependency graph and the variables used in the +computation. + +### Limitation 3 + +The user can not directly specify the parameter update rule for the +parameter server because the parameter server does not use the same +computation definition as the trainer. Instead, the update rule is +baked in the parameter server. The user can not specify the update +rule in the same way of specifying the trainer computation. + +This could be fixed by making the parameter server run the same +computation definition as the trainer. For a detailed explanation, +please +see +[Design Doc: Operation Graph Based Parameter Server](./dist_train.md) + +## Distributed Training Architecture + +The new distributed training architecture can address the above +limitations. Below is the illustration: + + + +The architecture includes major components: *PaddlePaddle Python*, +*PaddlePaddle converter* and *PaddlePaddle runtime*: + +### PaddlePaddle Python + +PaddlePaddle Python is the Python library that user's Python trainer +invoke to build the neural network topology, start training, etc. + +```Python +paddle.init() +input = paddle.op.recordIO("/home/data/mnist.recordio") # file stored on the cluster +img, label = input[0], input[1] +hidden = paddle.layer.fc(input=img, size=200, act=paddle.activation.Tanh()) +prediction = paddle.layer.fc(input=img, size=10, act=paddle.activation.Softmax()) +cost = paddle.layer.classification_cost(input=prediction, label=label) +optimizer = paddle.optimizer.SGD(cost, learning_rate=0.01) +session = paddle.session.NewRemote(num_trainer=3, num_ps=2, GPU_per_trainer=1) +for i in range(1000): + _, cost_val = session.eval(targets=[cost, optimizer]) + print cost_val +``` + +The code above is a typical Python trainer code, the neural network +topology is built using helper functions such as +`paddle.layer.fc`. The training is done by calling `session.eval` +iteratively. + +#### session.eval + +As shown in the graph, `session.eval` sends the IR and the evaluation +inputs/targets to the PaddlePaddle cluster for evaluation. The +targets can be any variable in the computation graph. When the target +is the `optimizer` variable, the neural network will be optimized +once. When the target is the `cost` variable, `session.eval` returns +the cost value. + +The Python `session` is a wrapper of the C++ `Session` class. For more +information about `Session`, please +see [Design Doc: Session](./session.md). + +### PaddlePaddle Converter + +PaddlePaddle converter automatically converts the IR in the request +(IR and evaluation inputs/targets) from PaddlePaddle Python to new +partitioned IRs and dispatch the new IRs and evaluation inputs/targets +to different PaddlePaddle runtimes. Below are the steps: + +1. Add `feed` OP that feeds the eval inputs, and `fetch` OP that + fetches the eval targets to the IR. + +1. Extract a new computation (sub)graph with `feed` and `fetch` OP as + the boundary. The runtime does not need to run the OP that is not + dependent by the `fetch` OP. + +1. Optimizes the computation graph. + +1. Place the OPs in the graph onto different devices on different + PaddlePaddle runtime according to a placement algorithm and device + constraint specified by the user. + +1. Partition the graph according to runtime boundaries and add `send` / + `recv` OP pair on the runtime boundaries. + +1. Dispatch the partitioned graph to different PaddlePaddle runtimes. + +1. PaddlePaddle runtimes with the `fetch` OP reports evaluation + results back to the converter, the convert reports the evaluation + results back to the PaddlePaddle Python. + +The output IRs will be cached to optimize the conversion latency. + + +#### Placement Algorithm + +Our first implementation will only support "trainer-parameter server" +placement: the parameters, initializers, and optimizers are placed on +the PaddlePaddle runtimes with the parameter server role. And +everything else will be placed on the PaddlePaddle runtimes with the +trainer role. This has the same functionality of our +"trainer-parameter server" architecture of PaddlePaddle v0.10.0, but +is more general and flexible. + +In the future, we will implement the general placement algorithm, +which makes placements according to the input IR, and a model of +device computation time and device communication time. Model +parallelism requires the general placement algorithm. + + +### PaddlePaddle Runtime + +The PaddlePaddle runtime owns multiple devices (e.g., CPUs, GPUs) and +runs the IR. The runtime does not need to do OP placement since it's +already done by the converter. + + +### Local Training Architecture + +The local training architecture will be the same as the distributed +training architecture, the differences are everything runs locally, +and there is just one PaddlePaddle runtime: + + + + +### Training Data + +In PaddlePaddle v0.10.0, training data is typically read +with [data reader](../reader/README.md) from Python. This approach is +no longer efficient when training distributedly since the Python +process no longer runs on the same node with the trainer processes, +the Python reader will need to read from the distributed filesystem +(assuming it has the access) and send to the trainers, doubling the +network traffic. + +When doing distributed training, the user can still use Python data +reader: the training data are sent with `session.eval`. However should +be used for debugging purpose only. The users are encouraged to use +the read data OPs. + + +## References: + +[1] [TensorFlow: Large-Scale Machine Learning on Heterogeneous Distributed Systems](https://static.googleusercontent.com/media/research.google.com/en//pubs/archive/45166.pdf) + +[2] [TensorFlow: A System for Large-Scale Machine Learning](https://www.usenix.org/system/files/conference/osdi16/osdi16-abadi.pdf) diff --git a/doc/design/ops/dist_train.md b/doc/design/refactor/parameter_server.md similarity index 100% rename from doc/design/ops/dist_train.md rename to doc/design/refactor/parameter_server.md diff --git a/doc/design/refactor/session.md b/doc/design/refactor/session.md new file mode 100644 index 0000000000000000000000000000000000000000..1d9a26683c14f54e3b5fe41675cd03b5620646b8 --- /dev/null +++ b/doc/design/refactor/session.md @@ -0,0 +1,180 @@ +# Design Doc: Session + +## Abstract + +The *session* object encapsulates the environment in which the +computation graph is executed. + +We will have the *local* session and *remote* session, they offer the +same [interface](#interface). The local session encapsulates the local +runtime environment and the remote session encapsulates the cluster +runtime environment. + +The local runtime environment contains: + +1. computation devices (i.e., CPU, GPU) handles, and +1. the [scope](../scope.md) which holds all variables. + +The remote runtime environment contains: + +1. computation devices (i.e., CPU and GPU on node 0, 1) in a cluster, + and +1. the distributed [scope](../scope.md) in a cluster which holds all + variables. + +The user can create a remote session on Paddle Cloud and evaluate the +computation graph with it. In this way, the user can control the +remote computation resource in a cluster from his local computer. + + +## Background + +The current design has an implicit global session in which +`paddle.eval()` is executed. The pain point is: + +Since the user is not able to explicitly switch between runtime +environments, the user cannot run a topology in two independent +environments. + +For example, in reinforcement learning, the user may want to have a +stale model for inference and a fresh model for training, and only +replace the stale model with the fresh model periodically. + +Furthermore, we have no concept that encapsulates a remote environment +that executes a computation graph. + +We need the session object to address above issues. + + +## Session + +A session is an object that owns the runtime environment. All +computations are executed through `session.eval()`. + + +### Interface + +```python +eval( + targets, + feed_dict=None, +) +``` + +Evaluates the target Operations or Variables in `targets`. + +- *targets*: the evaluation targets. Can be a single Operation or + Variable, or a list with the Operations or Variables as + elements. The value returned by `eval()` has the same shape as the + `target` argument. + + The PaddlePaddle program is represented by + the [ProgramDesc](../design/program.md), `eval()` will infer the + ProgramDesc from the given targets and run the PaddlePaddle + program. Please + see + [this graph](./distributed_architecture.md#local-training-architecture) for + the detailed illustration for the local session + and + [this graph](./distributed_architecture.md#distributed-training-architecture) for + the detailed illustration for the remote session. + +- *feed_dict*: a dictionary that contains the tensors which override + the edges of the computation graph. + + feed_dict not only can provide the input data, it can override any + OP's input as well: + + ```python + a = pd.constant(2.0, name="a") + b = pd.variable(name="b") + c = pd.mul(a,b) + sess.eval(targets=c, feed_dict={"b":3.0}) # returns 6.0 + ``` + +```python +close() +``` + +Closes the session and releases the scope that the session owns. + + +### Create a Local Session + +```python +session( + devices=None +) +``` + +Creates a new session. One session owns one global scope, so creating +multiple sessions will create different scopes. + +- *devices*: a single `string` or a list of `string` of device names, + the corresponding devices will be the computation devices for + `eval()`. If not specified, all available devices (e.g., all GPUs) + will be used. The user doesn't need to specify the CPU device since + it will be always used. Multiple sessions can use the same device. + + +#### Example + +```Python +a = paddle.constant(1.0) +b = paddle.constant(2.0) +c = a + b +sess = paddle.session(devices=["gpu:0", "gpu:1", "fpga:0"]) +sess.eval(c) +sess.close() +``` + +### Create a Remote Session + +```python +create_cloud_job( + name, + num_trainer, + mem_per_trainer, + gpu_per_trainer, + cpu_per_trainer, + num_ps, + mem_per_ps, + cpu_per_ps, +) +``` + +Creates a Paddle Cloud job. Fails if the job name exists. + +```python +get_cloud_job( + name +) +``` + +Gets a Paddle Cloud job. + +```python +remote_session( + job +) +``` + +- *job*: the Paddle Cloud job. + +#### Example + +```Python +reader = paddle.reader.recordio("/pfs/home/peter/mnist-train-*") # data stored on Paddle Cloud +image = reader.column(0) +label = reader.column(1) +fc1 = paddle.op.fc(image, size=256, act="sigmoid") +fc2 = paddle.op.fc(fc1, size=10, act="softmax") +cost = paddle.op.cross_entropy(fc2, label) +opt = paddle.optimizer.sgd(cost) + +job = paddle.create_cloud_job("test", 3, "1G", 1, 1, 2, "1G", 1) +sess = paddle.remote_ession(job) +for i in range(1000): + sess.eval(opt) +sess.close() +``` diff --git a/doc/design/refactor/src/compiler.graffle b/doc/design/refactor/src/compiler.graffle new file mode 100644 index 0000000000000000000000000000000000000000..8cc678fea3c820103e7ce81f7a5d625d6c1d92de Binary files /dev/null and b/doc/design/refactor/src/compiler.graffle differ diff --git a/doc/design/refactor/src/compiler.png b/doc/design/refactor/src/compiler.png new file mode 100644 index 0000000000000000000000000000000000000000..65d34f841afce9756def07dd8ecb9ca44e658bfe Binary files /dev/null and b/doc/design/refactor/src/compiler.png differ diff --git a/doc/design/ops/src/dist-graph.graffle b/doc/design/refactor/src/dist-graph.graffle similarity index 100% rename from doc/design/ops/src/dist-graph.graffle rename to doc/design/refactor/src/dist-graph.graffle diff --git a/doc/design/ops/src/dist-graph.png b/doc/design/refactor/src/dist-graph.png similarity index 100% rename from doc/design/ops/src/dist-graph.png rename to doc/design/refactor/src/dist-graph.png diff --git a/doc/design/refactor/src/distributed_architecture.graffle b/doc/design/refactor/src/distributed_architecture.graffle new file mode 100644 index 0000000000000000000000000000000000000000..f8496e57326c38de7468eb452a7713291d57653c Binary files /dev/null and b/doc/design/refactor/src/distributed_architecture.graffle differ diff --git a/doc/design/refactor/src/distributed_architecture.png b/doc/design/refactor/src/distributed_architecture.png new file mode 100644 index 0000000000000000000000000000000000000000..410c4510c6aab301dec95e6427fe80ac24e105fe Binary files /dev/null and b/doc/design/refactor/src/distributed_architecture.png differ diff --git a/doc/design/ops/src/local-graph.graffle b/doc/design/refactor/src/local-graph.graffle similarity index 100% rename from doc/design/ops/src/local-graph.graffle rename to doc/design/refactor/src/local-graph.graffle diff --git a/doc/design/ops/src/local-graph.png b/doc/design/refactor/src/local-graph.png similarity index 100% rename from doc/design/ops/src/local-graph.png rename to doc/design/refactor/src/local-graph.png diff --git a/doc/design/refactor/src/local_architecture.graffle b/doc/design/refactor/src/local_architecture.graffle new file mode 100644 index 0000000000000000000000000000000000000000..cc7783c45381f25ded0b898649322c81418ad317 Binary files /dev/null and b/doc/design/refactor/src/local_architecture.graffle differ diff --git a/doc/design/refactor/src/local_architecture.png b/doc/design/refactor/src/local_architecture.png new file mode 100644 index 0000000000000000000000000000000000000000..4b999538b7825c805292ee28b5e3256d5543bd09 Binary files /dev/null and b/doc/design/refactor/src/local_architecture.png differ diff --git a/doc/design/refactor/src/paddle-compile.graffle b/doc/design/refactor/src/paddle-compile.graffle new file mode 100644 index 0000000000000000000000000000000000000000..a6348cc3dbcaca923c6e794681b2edb85cb9f8f6 Binary files /dev/null and b/doc/design/refactor/src/paddle-compile.graffle differ diff --git a/doc/design/refactor/src/paddle-compile.png b/doc/design/refactor/src/paddle-compile.png new file mode 100644 index 0000000000000000000000000000000000000000..e0f13d551ac41afaec627a57dea79356464bf0bf Binary files /dev/null and b/doc/design/refactor/src/paddle-compile.png differ diff --git a/doc/design/refactorization.md b/doc/design/refactorization.md index ad801ca421ca31c84b0a6b0a18d1d625c87e0de5..f93d6155e1764386b01d2f0df3f141ab75cd55d4 100644 --- a/doc/design/refactorization.md +++ b/doc/design/refactorization.md @@ -1,40 +1,40 @@ # Design Doc: Refactorization Overview -The goal of refactorizaiton include: +The goals of refactoring include: -1. Make it easy for external contributors to write new elementory computaiton operations. -1. Make the codebase clean and readable. -1. Introduce a new design of computation representation -- a computation graph of operators and variables. -1. The graph representation helps implementing auto-scalable and auto fault recoverable distributed computing. +1. Making it easy for external contributors to write new elementary computation operations. +1. Making the codebase clean and readable. +1. Designing a new computation representation -- a computation graph of operators and variables. +1. Implementing auto-scalability and auto fault recoverable distributed computing with the help of computation graphs. ## Computation Graphs -1. PaddlePaddle represent the computation, training and inference of DL models, by computation graphs. +1. PaddlePaddle represents the computation, training and inference of Deep Learning models, by computation graphs. - 1. Please dig into [computation graphs](https://github.com/PaddlePaddle/Paddle/blob/develop/doc/design/graph.md) for a solid example. + 1. Please refer to [computation graphs](https://github.com/PaddlePaddle/Paddle/blob/develop/doc/design/graph.md) for a concrete example. -1. Users write Python programs to describe the graphs and run it (locally or remotely). +1. Users write Python programs to describe the graphs and run them (locally or remotely). 1. A graph is composed of *variables* and *operators*. -1. The description of graphs must be able to be serialized/deserialized, so it +1. The description of graphs must be serializable/deserializable, so that: - 1. could to be sent to the cloud for distributed execution, and - 1. be sent to clients for mobile or enterprise deployment. + 1. It can be sent to the cloud for distributed execution, and + 1. It can be sent to clients for mobile or enterprise deployment. -1. The Python program do +1. The Python program does two things - 1. *compilation*: runs a Python program to generate a protobuf message representation of the graph and send it to + 1. *Compilation* runs a Python program to generate a protobuf message representation of the graph and send it to 1. the C++ library `libpaddle.so` for local execution, 1. the master process of a distributed training job for training, or 1. the server process of a Kubernetes serving job for distributed serving. - 1. *execution*: according to the protobuf message, constructs instances of class `Variable` and `OperatorBase`, and run them. + 1. *Execution* executes the graph by constructing instances of class [`Variable`](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/framework/variable.h#L24) and [`OperatorBase`](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/framework/operator.h#L70), according to the protobuf message. -## Description and Realization +## Description and Realization of Computation Graph -At compile time, the Python program generates protobuf message representation of the graph, or the description of the graph. +At compile time, the Python program generates a protobuf message representation of the graph, or a description of the graph. -At runtime, the C++ program realizes the graph and run it. +At runtime, the C++ program realizes the graph and runs it. | | Representation (protobuf messages) | Realization (C++ class objects) | |---|---|---| @@ -42,30 +42,31 @@ At runtime, the C++ program realizes the graph and run it. |Operation|[OpDesc](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/framework/framework.proto#L35)|[Operator](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/framework/operator.h#L64)| |Block|BlockDesc|Block| -The word *graph* is exchangable with *block* in this document. A graph represent computation steps and local variables as a C++/Java program block, or a pair of { and }. +The word *graph* is interchangeable with *block* in this document. A graph consists of computation steps and local variables similar to a C++/Java program block, or a pair of parentheses(`{` and `}`). ## Compilation and Execution -1. Run an applicaton Python program to describe the graph. In particular, +1. Run a Python program to describe the graph. In particular, the Python application program does the following: - 1. create VarDesc to represent local/intermediate variables, - 1. create operators and set attributes, - 1. validate attribute values, - 1. inference the type and the shape of variables, - 1. plan for memory-reuse for variables, - 1. generate backward and optimization part of the Graph. - 1. possiblly split the graph for distributed training. + 1. Create `VarDesc` to represent local/intermediate variables, + 1. Create operators and set attributes, + 1. Validate attribute values, + 1. Infer the type and the shape of variables, + 1. Plan memory-reuse for variables, + 1. Generate the backward graph + 1. Add optimization operators to the computation graph. + 1. Optionally, split the graph for distributed training. -1. The invocation of `train` or `infer` in the application Python program: +1. The invocation of `train` or [`infer`](https://github.com/PaddlePaddle/Paddle/blob/develop/python/paddle/v2/inference.py#L108) methods in the Python program does the following: - 1. create a new Scope instance in the [scope hierarchy](https://github.com/PaddlePaddle/Paddle/blob/develop/doc/design/scope.md) for each run of a block, + 1. Create a new Scope instance in the [scope hierarchy](https://github.com/PaddlePaddle/Paddle/blob/develop/doc/design/scope.md) for each run of a block, 1. realize local variables defined in the BlockDesc message in the new scope, 1. a scope is similar to the stack frame in programming languages, - 1. create an instance of class `Block`, in which, + 1. Create an instance of class `Block`, in which, 1. realize operators in the BlockDesc message, - 1. run the Block by calling + 1. Run the Block by calling 1. `Block::Eval(vector* targets)` for forward and backward computations, or 1. `Block::Eval(vector* targets)` for optimization. @@ -76,14 +77,14 @@ The word *graph* is exchangable with *block* in this document. A graph represen Compile Time -> IR -> Runtime ``` -### Benefit +### Benefits of IR - Optimization ```text Compile Time -> IR -> Optimized IR -> Runtime ``` -- Send automatically partitioned IR to different nodes. - - Automatic data parallel +- Automatically send partitioned IR to different nodes. + - Automatic Data Parallelism ```text Compile Time |-> Single GPU IR @@ -92,7 +93,7 @@ Compile Time -> IR -> Runtime |-> Node-1 (runs trainer-IR-1) |-> Node-2 (runs pserver-IR) ``` - - Automatic model parallel (planned for future) + - Automatic Model Parallelism (planned for future) --- @@ -105,10 +106,10 @@ Compile Time -> IR -> Runtime # Operator ![class_diagram](http://api.paddlepaddle.org/graphviz?dot=https://gist.githubusercontent.com/reyoung/53df507f6749762675dff3e7ce53372f/raw/dd598e8f1976f5759f58af5e5ef94738a6b2e661/op.dot) -* `Operator` is the fundamental building block as the user interface. - * Operator stores input/output variable name, and attributes. - * The `InferShape` interface is used to infer output variable shapes by its input shapes. - * Use `Run` to compute `input variables` to `output variables`. +* `Operator` is the fundamental building block of the user interface. + * Operator stores input/output variable names and attributes. + * The `InferShape` interface is used to infer the shape of the output variables based on the shapes of the input variables. + * Use `Run` to compute the `output` variables from the `input` variables. --- @@ -126,30 +127,29 @@ Compile Time -> IR -> Runtime # Why separate Kernel and Operator * Separate GPU and CPU code. - * Make Paddle can run without GPU. -* Make one operator (which is user interface) can contain many implementations. - * Same mul op, different FP16, FP32 Kernel. different MKL, eigen kernel. + * Make Paddle capable of running without GPU. +* Make one operator (which is a user interface) and create many implementations. + * For example, same multiplication op can have different implementations kernels such as FP16 kernel, FP32 kernel, MKL, eigen kernel. --- # Libraries for Kernel development * `Eigen::Tensor` contains basic math and element-wise functions. * Note that `Eigen::Tensor` has broadcast implementation. - * Limit number of `tensor.device(dev) = ` in your code. -* `thrust::tranform` and `std::transform`. - * `thrust` has the same API as C++ standard library. Using `transform` can quickly implement a customized elementwise kernel. - * `thrust` has more complex API, like `scan`, `reduce`, `reduce_by_key`. + * Limit the number of `tensor.device(dev) = ` in your code. +* `thrust::transform` and `std::transform`. + * `thrust` has the same API as C++ standard library. Using `transform`, one can quickly implement customized element-wise kernels. + * `thrust`, in addition, supports more complex APIs, like `scan`, `reduce`, `reduce_by_key`. * Hand-writing `GPUKernel` and `CPU` code - * Do not write `.h`. CPU Kernel should be in `.cc`. GPU kernel should be in `.cu`. (`GCC` cannot compile GPU code.) + * Do not write in header (`.h`) files. CPU Kernel should be in cpp source (`.cc`) and GPU kernels should be in cuda (`.cu`) files. (GCC cannot compile GPU code.) --- -# Operator Register +# Operator Registration -## Why register is necessary? +## Why is registration necessary? We need a method to build mappings between Op type names and Op classes. -## How to do the register? - -Maintain a map, whose key is the type name and value is corresponding Op constructor. +## How is registration implemented? +Maintaining a map, whose key is the type name and the value is the corresponding Op constructor. --- # The Registry Map @@ -169,7 +169,7 @@ Maintain a map, whose key is the type name and value is corresponding Op constru # Related Concepts ### Op_Maker -It's constructor takes `proto` and `checker`. They are compeleted during Op_Maker's construction. ([ScaleOpMaker](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/operators/scale_op.cc#L37)) +It's constructor takes `proto` and `checker`. They are completed during Op_Maker's construction. ([ScaleOpMaker](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/operators/scale_op.cc#L37)) ### Register Macros ```cpp @@ -177,34 +177,30 @@ REGISTER_OP(op_type, op_class, op_maker_class, grad_op_type, grad_op_class) REGISTER_OP_WITHOUT_GRADIENT(op_type, op_class, op_maker_class) ``` -### `USE` Macros -make sure the registration process is executed and linked. - --- -# Register Process -1. Write Op class, as well as its gradient Op class if there is. -2. Write Op maker class. In the constructor, describe its inputs, outputs, and attributes. -3. Invoke macro `REGISTER_OP`. The macro will - 1. call maker class to complete `proto` and `checker` - 2. with the completed `proto` and `checker`, build a new key-value pair in the `OpInfoMap` - -4. Invoke `USE` macro in where the Op is used to make sure it is linked. +# Registration Process +1. Write an Op class and its gradient Op class, if required. +2. Write an Op maker class. In the constructor of this class, describe the inputs, outputs and attributes of the operator. +3. Invoke the macro `REGISTER_OP`. This macro will + 1. Call maker class to complete `proto` and `checker` + 2. Using the completed `proto` and `checker`, it will add a new key-value pair to the `OpInfoMap` --- # Backward Module (1/2) ### Create Backward Operator -- Mapping from forwarding Op to backward Op +- Mapping from forward Op to backward Op ![backward](https://gist.githubusercontent.com/dzhwinter/a6fbd4623ee76c459f7f94591fd1abf0/raw/61026ab6e518e66bde66a889bc42557a1fccff33/backward.png) --- # Backward Module (2/2) ### Build Backward Network -- **Input** graph of forwarding operators -- **Output** graph of backward operators -- **corner case in construction** - - shared variable => insert `Add` operator - - no gradient => insert `fill_zero_grad` operator - - recursive netOp => call `Backward` recursively +- **Input**: a graph of forward operators +- **Output**: a graph of backward operators +- **Corner cases in construction** + - Shared Variables => insert an `Add` operator to combine gradients + - No Gradient => insert a `fill_zero_grad` operator + - Recursive NetOp => call `Backward` recursively + - RNN Op => recursively call `Backward` on stepnet - RNN Op => recursively call `Backward` on stepnet @@ -213,41 +209,41 @@ make sure the registration process is executed and linked. * `Tensor` is an n-dimension array with type. * Only dims and data pointers are stored in `Tensor`. - * All operators on `Tensor` is written in `Operator` or global functions. - * variable length Tensor design [LoDTensor](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/framework/lod_tensor.md) -* `Variable` is the inputs and outputs of an operator. Not just `Tensor`. - * step_scopes in RNN is a variable and not a tensor. -* `Scope` is where variables store at. - * map - * `Scope` has a hierarchical structure. The local scope can get variable from its parent scope. + * All operations on `Tensor` are written in `Operator` or global functions. + * Variable length Tensor design [LoDTensor](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/framework/lod_tensor.md) +* `Variable` instances are the inputs and the outputs of an operator, not just `Tensor`. + * `step_scopes` in RNN is a variable and not a tensor. +* `Scope` is where variables are stored. + * map + * `Scope` has a hierarchical structure. The local scope can get variables from its parent scope. --- # Block (in design) -## the difference with original RNNOp -- as an operator is more intuitive than `RNNOp`, -- offers new interface `Eval(targets)` to deduce the minimal block to `Run`, -- fits the compile-time/ runtime separation design. - - during the compilation, `SymbolTable` stores `VarDesc`s and `OpDesc`s and serialize to a `BlockDesc` - - when graph executes, a Block with `BlockDesc` passed in creates `Op` and `Var` then `Run` +## the difference between original RNNOp and Block +- As an operator is more intuitive than `RNNOp`, +- Offers a new interface `Eval(targets)` to deduce the minimal block to `Run`, +- Fits the compile-time/ runtime separation design paradigm. + - During the compilation, `SymbolTable` stores `VarDesc`s and `OpDesc`s and serialize to a `BlockDesc` + - When graph executes, a Block with `BlockDesc` is passed. It then creates `Op` and `Var` instances and then invokes `Run`. --- # Milestone -- take Paddle/books as the main line, the requirement of the models motivates framework refactoring, -- model migration - - framework development gives **priority support** to model migration, for example, +- Take Paddle/books as the main line, the requirement of the models motivates framework refactoring, +- Model migration + - Framework development gives **priority support** to model migration, for example, - the MNIST demo needs a Python interface, - the RNN models require the framework to support `LoDTensor`. - - determine some timelines, - - heavily-relied Ops need to be migrated first, - - different models can be migrated parallelly. -- improve the framework at the same time -- accept imperfection, concentrated on solving the specific problem at the right price. + - Determine some timelines, + - Frequently used Ops need to be migrated first, + - Different models can be migrated in parallel. +- Improve the framework at the same time +- Accept imperfection, concentrate on solving the specific problem at the right price. --- # Control the migration quality -- compare the performance of migrated models with old ones. -- follow google C style -- build the automatic workflow of generating Python/C++ documentations - - the documentation of layers and ops should be written inside the code - - take the documentation quality into account when doing PR - - preview the documentations, read and improve them from users' perspective +- Compare the performance of migrated models with old ones. +- Follow the google C++ style guide. +- Build the automatic workflow of generating Python/C++ documentations. + - The documentation of layers and ops should be written inside the code. + - Take the documentation quality into account when submitting pull requests. + - Preview the documentations, read and improve them from a user's perspective. diff --git a/doc/design/register_grad_op.md b/doc/design/register_grad_op.md new file mode 100644 index 0000000000000000000000000000000000000000..8d973eb53178c3e889c845144553a453e11f067c --- /dev/null +++ b/doc/design/register_grad_op.md @@ -0,0 +1,92 @@ +# Design Doc: Gradient Operators Registration + + +## The Problem Posed + +Currently, for each C++ operator class definition, a *gradient operator creator* function is registered, which takes as input a C++ operator instance and returns the corresponding gradient operator instance. + +However, we noticed two problems with the current design: + +1. As we decided to separate the *compilation* and the *execution* phases, we need to change the creator to take an `OpDesc` protobuf message in a `ProgramDesc` and inserts corresponding `OpDesc` messages into the `ProgramDesc` message. + +1. For some operators, the gradient computation can be written in terms of existing operators. For example, the gradient of *minus* operator consists of two operators -- an *identity* operator followed by a *scale* operator. Hence the registration mechanism needs to support mapping from an operator to a set of operators for the gradient computation. + +## The Current Implementation + +Instances of the C++ class `OpInfo` are stored an associative map whose key is the operator type. The `grad_op_type` indicates the associated gradient operator type. An operator can create the gradient operator by invoking `OpInfo::creator_` of the gradient operator. The pseudo code is as follows + +```cpp +struct OpInfo { + std::function creator_; + std::string grad_op_type_; + ... +}; + +map OpInfoMap; + +OperatorBase* CreateGradientOperator(const OperatorBase& op) { + return OpInfoMap.at(op.Type()).creator_(...); +} +``` + +## Proposed Solution + +The mapping relationship between an operator and its gradient operators is a function. The interface of this function is: + +```cpp +// (OpDesc) --> vector +std::function(const OpDescBind&)>; +``` + +The function takes an `OpDescBind` of the forward operator and returns one or many gradient operator descriptions. `OpDescBind` is a C++ wrapper for the protobuf message `OpDesc` for rapid manipulation of `OpDesc`. + +The `GradOpDescMaker` will be registered in `OpInfo` and will replace the `grad_op_type_` field. The `OpInfo` should look like + +```cpp +struct OpInfo { + std::function>(const OpDescBind&)> grad_op_maker_; + ... +}; +``` + +The `grad_op_maker_ ` is a `nullptr` if the operator does not have any associated gradient operators. + +We propose a base class called `GradOpDescMakerBase` to let operator developers generate `Gradient Operators` easily. The public interface of that class is + +```cpp +class GradOpDescMakerBase { +public: + GradOpDescMakerBase(const OpDescBind& ); + virtual std::vector> operator()()const = 0; +}; +``` + +We can convert `GradOpDescMakerBase` to `std::function>(const OpDescBind&)>` by + +```cpp +using GradOpMaker = ...; +std::function(const OpDescBind&)> func; +func = [] (const OpDescBind& fwd_op) { + GradOpMaker maker(fwd_op); + return maker(); +}; +``` + +We can write many helper functions since the `GradOpDescMakerBase` is a class now. The basic helper functions get the variables of `Input`, `Output`, `InputGradient` and `OutputGradient` in the forwarding operator. + +We should change register macros at the same time. In the current solution, there is no difference between forwarding operators and backward operators. So `REGISTER_OP` just register one operator. If the `REGISTER_OPERATOR ` contains `OpProtoAndCheckerMaker` and `GradOpDescMaker`, we just list them in the same macro. It can be done by a macro contains `__VA_ARGS__`. + +The user interface should be + +```cpp +vector MinusOpGradMaker(OpDesc) {...} +REGISTER_OPERATOR(minus, MinusOp, MinusOpProtoAndCheckerMaker, SumOpGradMaker); +// Developers can still manually implement gradient operator. +REGISTER_OPERATOR(minus_grad, MinusGradOp); +``` + +The interface of current `REGISTER_OP` macro could not be changed. In `REGISTER_OP`, it will invoke `REGISTER_OPERATOR` two times and generate GradOpDescMaker inside. + +```cpp +REGISTER_OP(minus, MinusOp, MinusOpProtoAndCheckerMaker, minus_grad, MinusGradOp); +``` diff --git a/doc/design/regularization.md b/doc/design/regularization.md new file mode 100644 index 0000000000000000000000000000000000000000..21280ac898feb4dd5e5a5d9e88d121e856850f0b --- /dev/null +++ b/doc/design/regularization.md @@ -0,0 +1,72 @@ +# Regularization in PaddlePaddle + +## Introduction to Regularization +A central problem in machine learning is how to design an algorithm that will perform well not just on the training data, but also on new data. A frequently faced problem is the problem of **overfitting**, where the model does not make reliable predictions on new unseen data. **Regularization** is the process of introducing additional information in order to prevent overfitting. This is usually done by adding extra penalties to the loss function that restricts the parameter spaces that an optimization algorithm can explore. + +### Parameter Norm Penalties +Most common regularization approaches in deep learning are based on limiting the capacity of the models by adding a parameter norm penalty to the objective function `J`. This is given as follows: + +
+ +The parameter `alpha` is a hyperparameter that weights the relative contribution of the norm penalty term, `omega`, relative to the standard objective function `J`. + +The most commonly used norm penalties are the L2 norm penalty and the L1 norm penalty. These are given as follows: + +##### L2 Regularization: +
+ +##### L1 Regularization +
+ +A much more detailed mathematical background of regularization can be found [here](http://www.deeplearningbook.org/contents/regularization.html). + +## Regularization Survey + +A detailed survey of regularization in various deep learning frameworks can be found [here](https://github.com/PaddlePaddle/Paddle/wiki/Regularization-Survey). + +## Proposal for Regularization in PaddlePaddle + +### Low-Level implementation + +In the new design, we propose to create new operations for regularization. For now, we can add 2 ops that correspond to the most frequently used regularizations: +- L2_regularization_op +- L1_regularization_op + +These ops can be like any other ops with their own CPU/GPU implementations either using Eigen or separate CPU and GPU kernels. As the initial implementation, we can implement their kernels using Eigen following the abstraction pattern implemented for [Activation Ops](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/operators/accuracy_op.h). This abstraction pattern can make it very easy to implement new regularization schemes other than L1 and L2 norm penalties. + +The idea of building ops for regularization is in sync with the refactored Paddle philosophy of using operators to represent any computation unit. The way these ops will be added to the computation graph, will be decided by the [layer functions](https://github.com/PaddlePaddle/Paddle/blob/develop/doc/design/python_api.md#layer-function) in Python API. + +### Computation Graph + +Below is an example of a really simple feed forward neural network. + +
+ +The Python API will modify this computation graph to add regularization operators. The modified computation graph will look as follows: + +
+    +### Python API implementation for Regularization + +Using the low level ops, `L2_regularization_op` and `L1_regularization_op`, any user can add regularization to their computation graphs. However, this will require a lot of lines of code and we should design Python APIs that support regularization. An example of such an API can be seen in [Keras](https://keras.io/regularizers/). As per the PaddlePaddle [Python API design](https://github.com/PaddlePaddle/Paddle/blob/develop/doc/design/python_api.md), the layer functions are responsible for creating operators, operator parameters and variables. Since regularization is a property of parameters, it makes sense to create these in the layer functions. + +#### Creation of Regularization ops +There are two possibilities for creating the regularization ops: +1. We create these ops immediately while building the computation graph. +2. We add these ops in a lazy manner, just before the backward, similar to the way the optimization ops are added. + +The proposal is to add these ops in a lazy manner just before the backward pass. + +#### Storage of Regularization attributes + +Since we want to create the regularization ops in a lazy manner, the regularization attributes (type of regularization and weight of regularization penalty) can be stored as attributes of the [`Parameter`](https://github.com/PaddlePaddle/Paddle/blob/develop/python/paddle/v2/framework/framework.py#L421) class. This is because regularization is a property of the parameters and storing regularization properties with Parameters also allows for shared parameters. + +#### High-level API + +In PaddlePaddle Python API, users will primarily rely on [layer functions](https://github.com/PaddlePaddle/Paddle/blob/develop/doc/design/python_api.md#layer-function) to create neural network layers. Hence, we also need to provide regularization functionality in layer functions. The design of these APIs can be postponed for later right now. A good reference for these APIs can be found in [Keras](https://keras.io/regularizers/) and also by looking at Tensorflow in [`tf.contrib.layers`](https://www.tensorflow.org/api_guides/python/contrib.layers). + + + + + + diff --git a/doc/design/scope.md b/doc/design/scope.md index b1f9bb4378eb5ec6926f1e53f7c1f4fd5674064c..4da76eebb74abcd26ec2b8671399e6bc4fb58574 100644 --- a/doc/design/scope.md +++ b/doc/design/scope.md @@ -37,7 +37,7 @@ Scope is an association of a name to variable. All variables belong to `Scope`. ```cpp class Scope { public: - Variable* NewVar(const std::string& name); + Variable* Var(const std::string& name); const Variable* FindVar(const std::string& name) const; private: @@ -98,7 +98,7 @@ class Scope { Variable* FindVar(const std::string& name) const; // return if already contains same name variable. - Variable* NewVar(const std::string& name); + Variable* Var(const std::string& name); private: std::shared_ptr parent_; @@ -107,7 +107,7 @@ class Scope { ``` ## Only scope can create a variable -To ensure `only scope can create a variable`, we should mark `Variable`'s constructor as a private member function, and Scope is a friend class of Variable. And then only `NewVar` can construct `Variable`. +To ensure `only scope can create a variable`, we should mark `Variable`'s constructor as a private member function, and Scope is a friend class of Variable. And then only `Var` can construct `Variable`. ## When scope destroyed, all variables inside this scope should be destroyed together @@ -121,4 +121,4 @@ Also, as the parent scope is a `shared_ptr`, we can only `Create()` a scope shar ## Orthogonal interface -`FindVar` will return `nullptr` when `name` is not found. It can be used as `Contains` method. `NewVar` will return an `Error` when there is a name conflict locally. Combine `FindVar` and `NewVar`, we can implement `NewVar` easily. +`FindVar` will return `nullptr` when `name` is not found. It can be used as `Contains` method. `Var` will return an `Error` when there is a name conflict locally. Combine `FindVar` and `Var`, we can implement `Var` easily. diff --git a/doc/design/selected_rows.md b/doc/design/selected_rows.md new file mode 100644 index 0000000000000000000000000000000000000000..1a98839a957612b91b2276b58818623ecc62d1d5 --- /dev/null +++ b/doc/design/selected_rows.md @@ -0,0 +1,74 @@ +# Design Doc: Selected Rows + +`SelectedRows` is a type of sparse tensor data type, which is designed to support `embedding` operators. The gradient of embedding table is a sparse tensor. Only a few rows are non-zero values in this tensor. It is straight-forward to represent a sparse tensor by the following sparse tensor data structure: + +```cpp +class SelectedRows { + private: + vector rows_; + Tensor value_; + int height_; +}; +``` + +The field `height_` is the first dimension of `SelectedRows`. The `rows` are the indices of the non-zero rows of `SelectedRows`. The `value_` field is an N-dim tensor of shape `[rows.size() /* NUM_ROWS */, ...]`, which supplies values for each row. The dimension of `SelectedRows` satisfies `[height_] + value_.shape[1:]`. + +Suppose that a SelectedRows-typed variable `x` has many rows, but only two of them have values -- row 73 is `[1, 2]` and row 84 is `[3, 4]`, the `SelectedRows` representation would be: + +``` +x = SelectedRow { + rows = [73, 84], + value = [[1, 2], [3,4]] +} +``` + + +## SelectedRows in Protobuf + +`SelectedRows` is a type of `Variable`. `VarDesc` in protobuf should describe the `SelectedRows` information. Only the tensor dimension of a `SelectedRows` will be described in compile-time because the `rows_` and `value_` are dependent on the training data. +So we use `TensorDesc` to unify `data_type` and `dims`. A LodTensorDesc contains a `TensorDesc` and `lod_level`. The description of `SelectedRows` is a Tensor description. + +```proto +message TensorDesc { + required DataType data_type = 1; + repeated int64 dims = 2; // [UNK, 640, 480] is saved as [-1, 640, 480] +} + +message LodTensorDesc { + required TensorDesc tensor = 1; + optional int lod_level = 2; +} + +message VarDesc { + required string name = 1; + enum VarType { + LOD_TENSOR = 0; + SELECTED_ROWS = 1; + } + required VarType type = 2; + optional LodTensorDesc lod_desc = 3; + optional TensorDesc selected_rows_desc = 4; + optional bool persistable = 5 [ default = false ]; +} +``` + +## InferShape for Selected Rows + +Just like `LoD` information, `InferShape` method will infer the output tensor type as well. The operator should decide whether its output is a `SelectedRows` or `Dense` tensor. + +For example, the gradient operator of `TableLookup` will always generate `SelectedRows`. Its `InferShape` method should be like following + +```cpp +void TableLookupGrad::InferShape(context) { + ... + context.SetDataType("Embedding.Grad", kSelectedRows); +} +``` + + +## Sparse Operators + +There are several operators that need to be written to support `SelectedRows`. These are: + +1. Operators which generate `SelectedRows` gradient. e.g. Gradient of `TableLookupOp`. +2. Optimize operators which support `SelectedRows` gradient. e.g. `SGD` or `AdaGrad` for `SelectedRows`. However, there should be only one `SGD` operator. `OpWithKernel::Run` should select a suitable kernel for both `dense` tensor or `SelectedRows`. diff --git a/doc/design/tensor_array.md b/doc/design/tensor_array.md new file mode 100644 index 0000000000000000000000000000000000000000..37e4f7b90f94fa3eb015e733999cd84c96b2239c --- /dev/null +++ b/doc/design/tensor_array.md @@ -0,0 +1,271 @@ +# Design for TensorArray +This design doc presents the necessity of a new C++ class `TensorArray`. +In addition to the very simple C++ implementation + +```c++ +class TensorArray { + public: + explicit TensorArray(const LoDTensor&); + explicit TensorArray(size_t size); + + private: + vector values_; +}; +``` + +We also need to expose it to PaddlePaddle's Python API, +because users would want to use it with our very flexible operators `WhileLoop`. +An example for a RNN based on dynamic operators is + +```python +input = pd.data(...) +num_steps = Var(12) + +TensorArray states(size=num_steps) +TensorArray step_inputs(unstack_from=input) +TensorArray step_outputs(size=num_steps) + +W = Tensor(...) +U = Tensor(...) +default_state = some_op() + +step = Var(1) + +wloop = paddle.create_whileloop(loop_vars=[step]) +with wloop.frame(): + wloop.break_if(pd.equal(step, num_steps) + pre_state = states.read(step-1, default_state) + step_input = step_inputs.read(step) + state = pd.sigmoid(pd.matmul(U, pre_state) + pd.matmul(W, step_input)) + states.write(step, state) + step_outputs.write(step, state) # output state + step.update(state+1) + +output = step_outputs.stack() +``` + +## Background +Steps are one of the core concepts of RNN. In each time step of RNN, there should be several input segments, states, and output segments; all these components act like arrays, for example, call `states[step_id]` will get the state in `step_id`th time step. + +An RNN can be implemented with the following pseudocode + +```c++ +Array states; +Array input_segments; +Array output_segments; +Parameter W, U; + +step = 1 +seq_len = 12 +while_loop { + if (step == seq_len) break; + states[step] = sigmoid(W * states[step-1] + U * input_segments[step]); + output_segments[step] = states[step] // take state as output + step++; +} +``` +According to the [RNN roadmap](https://github.com/PaddlePaddle/Paddle/issues/4561), there are several different RNNs that PaddlePaddle will eventually support. + +Currently, the basic RNN implementation supported by PaddlePaddle is the `recurrent_op` which takes tensors as input and splits them into `input_segments`. + + +Since a tensor cannot store variable-length sequences directly, PaddlePaddle implements the tensor with level of details (`LoDTensor` for short). +Segmenting the `LoDTensor` is much more complicated than splitting a tensor, that makes it necessary to refactor the `recurrent_op` with `LoDTensor` segmenting support. + +As the next step in RNN support, `dynamic_recurrent_op` should be introduced to handle inputs with variable-length sequences. + +The implementation is similar to `recurrent_op`. +The key difference is the way **the original input `LoDTensors` and outupts are split to get the `input_segments` and the `output_segments`.** + + +Though it can't be built over `recurrent_op` or `dynamic_recurrent_op` directly, +the logic behind splitting a tensor or a LoD tensor into `input_segments` remains the same. + +## Why `TensorArray` +The logic behind splitting the inputs to segments, states and outputs is similar and can be shared in a seperate module. + +The array of `states`, `input_segments` and `output_segments` would be exposed to users when writing a dynamic RNN model similar to the above pseudo codes. + +So there should be an array-like container, which can store the segments of a tensor or LoD tensor. + +**This container can store an array of tensors and provides several methods to split a tensor or a LoD tensor** . +This is where the notion of `TensorArray` comes from. + +## Introduce TensorArray to uniform all the three RNNs +TensorArray as a new concept is borrowed from TensorFlow, +it is meant to be used with dynamic iteration primitives such as `while_loop` and `map_fn`. + +This concept can be used to support our new design of dynamic operations, and help to refactor some existing variant-sentence-related layers, +such as `recurrent_op`, `RecurrentGradientMachine`. + +In [our design for dynamic RNN](https://github.com/PaddlePaddle/Paddle/pull/4401), +`TensorArray` is used to segment inputs and store states in all time steps. +By providing some methods similar to a C++ array, +the definition of some state-based dynamic models such as RNN can be more natural and highly flexible. + +## Dynamic-operations on TensorArray + +`TensorArray` will be used directly when defining dynamic models, so some operators listed below should be implemented + +```python +# several helper operators for TensorArray +def tensor_array_stack(ta, tensor): + ''' + get a tensor array `ta`, return a packed `tensor`. + ''' + pass + +def tensor_array_unstack(tensor, ta): + ''' + get a `tensor`, unstack it and get a tensor array `ta`. + ''' + pass + +def tensor_array_write(ta, index, tensor, data_shared): + ''' + get a `tensor` and a scalar tensor `index`, write `tensor` into index-th + value of the tensor array `ta`. + `data_shared` is an attribute that specifies whether to copy or reference the tensors. + ''' + pass + +def tensor_array_read(ta, index, tensor): + ''' + get a tensor array `ta`, a scalar tensor `index`, read the index-th value of + `ta` and return as the `tensor`. + ''' + pass + +def tensor_array_size(ta, tensor): + ''' + get a tensor array `ta`, return the size of `ta` and return as the scalar `tensor`. + ''' + pass +``` + +It is trivial for users to use so many low-level operators, so some helper methods should be proposed in python wrapper to make `TensorArray` easier to use, +for example + +```python +class TensorArray: + def __init__(self, name): + self.name = name + self.desc = TensorArrayDesc() + + def stack(self, name=None): + ''' + Pack the values in a `TensorArray` into a tensor with rank one higher + than each tensor in `values`. + `stack` can be used to split tensor into time steps for RNN or whileloop. + + @name: str + the name of the variable to output. + ''' + tensor = Var(name) + tensor_array_stack(self.name, tensor) + return tensor + + def unstack(self, input): + ''' + Unpacks the given dimension of a rank-`R` tensor into rank-`(R-1)` tensors. + `unstack` can be used to concatenate all the time steps for RNN or whileloop. + + @input: str + the name of input tensor + ''' + tensor_array_unstack(tensor, self.name) + + def write(self, index, value, data_shared=True): + ''' + Write value into index of the TensorArray. + If `data_shared` is set to True, than the index-th value in TensorArray will + be shared with the tensor passed in. + + @index: str + name of a scalar tensor + @value: str + name of a tensor + @data_shared: bool + ''' + tensor_array_write(self.name, index, value, data_shared) + + def read(self, index, output): + ''' + Read the value at location `index` in the `TensorArray`. + + @index: str + name of a scalar tensor + @output: + name of a output variable + ''' + tensor_array_read(self.name, index, output) + + + def size(self, output): + ''' + Return the number of values. + + @output: str + name of a scalar tensor + ''' + tensor_array_size(self.name, output) +``` + +## LoDTensor-related Supports +The `RecurrentGradientMachine` in Paddle serves as a flexible RNN layer; it takes varience-length sequences as input, and output sequences too. + +Since each step of RNN can only take a tensor-represented batch of data as input, +some preprocess should be taken on the inputs such as sorting the sentences by their length in descending order and cut each word and pack to new batches. + +Such cut-like operations can be embedded into `TensorArray` as general methods called `unpack` and `pack`, +these two operations are similar to `stack` and `unstack` except that they operate on variable-length sequences formated as a LoD tensor rather than a tensor. + +Some definitions are like + +```python +def unpack(level): + ''' + Split LodTensor in some `level` and generate batches, if set `sort_by_length`, + will sort by length. + + Returns: + - a new `TensorArray`, whose values are LodTensors and represents batches + of data. + - an int32 Tensor, which stores the map from the new batch's indices to + original LoDTensor + ''' + pass + +def pack(level, indices_map): + ''' + Recover the original LoD-arranged LoDTensor with the values in a `TensorArray` + and `level` and `indices_map`. + ''' + pass +``` + +With these two methods, a varience-length sentence supported RNN can be implemented like + +```c++ +// input is the varient-length data +LodTensor sentence_input(xxx); +TensorArray ta; +Tensor indice_map; +Tensor boot_state = xxx; // to initialize rnn's first state +TensorArray::unpack(input, 1/*level*/, true/*sort_by_length*/, &ta, &indice_map); +TessorArray step_outputs; +TensorArray states; + +for (int step = 0; step = ta.size(); step++) { + auto state = states.read(step); + // rnnstep is a function which acts like a step of RNN + auto step_input = ta.read(step); + auto step_output = rnnstep(step_input, state); + step_outputs.write(step_output, true/*data_shared*/); +} + +// rnn_output is the final output of an rnn +LoDTensor rnn_output = ta.pack(ta, indice_map); +``` +the code above shows that by embedding the LoDTensor-related preprocess operations into `TensorArray`, +the implementation of a RNN that supports varient-length sentences is far more concise than `RecurrentGradientMachine` because the latter mixes all the codes together, hard to read and extend. diff --git a/doc/design/test.dot b/doc/design/test.dot new file mode 100644 index 0000000000000000000000000000000000000000..62c69b8fc8010a26a54a6ee8ef1488aad94d747a --- /dev/null +++ b/doc/design/test.dot @@ -0,0 +1,35 @@ + +digraph Test { + z -> generator -> G_img; + G_img -> discriminator -> D_f -> d_loss_f; + label0 -> d_loss_f -> d_loss; + + img -> discriminator -> D_t -> d_loss_t; + label1 -> d_loss_t -> d_loss; + + d_loss -> d_loss_t[color=red, style=dashed]; + d_loss -> d_loss_f[color=red, style=dashed]; + d_loss_t -> D_t[color=red, style=dashed]; + d_loss_f -> D_f[color=red, style=dashed]; + D_t -> discriminator[color=red, style=dashed]; + D_f -> discriminator[color=red, style=dashed]; + + D_f -> g_loss; + label2 -> g_loss; + + g_loss -> D_f[color=green, style=dashed]; + D_f -> discriminator[color=green, style=dashed]; + discriminator -> G_img[color=green, style=dashed]; + G_img -> generator[color=green, style=dashed]; + + discriminator [color=red, shape=box]; + generator [color=green, shape=box]; + z [shape=diamond]; + img [shape=diamond]; + label0 [shape=diamond]; + label1 [shape=diamond]; + label2 [shape=diamond]; + + d_loss [color=red]; + g_loss [color=green]; +} diff --git a/doc/design/test.dot.png b/doc/design/test.dot.png new file mode 100644 index 0000000000000000000000000000000000000000..4e121a40b9f7b2232d7cdda315bad15926446f55 Binary files /dev/null and b/doc/design/test.dot.png differ diff --git a/doc/design/var_desc.md b/doc/design/var_desc.md index bfbbdd0578ebc69ea4b49ade9b041573a9e9ad55..0b2958c1b10ef6a6ce51aa75f61e15a7f2d94b3f 100644 --- a/doc/design/var_desc.md +++ b/doc/design/var_desc.md @@ -16,16 +16,23 @@ The computation graph is constructed by Data Node and Operation Node. The concep ## Definition of VarDesc -A VarDesc should have a name and value, in PaddlePaddle, the value will always be a tensor. Since we use LoDTensor most of the time. We add a LoDTesnorDesc to represent it. +A VarDesc should have a name, and value. The are two kinds of variable type in compile time, they are `LoDTensor` and `SelectedRows`. ```proto message VarDesc { required string name = 1; - optional LoDTensorDesc lod_tensor = 2; + enum VarType { + LOD_TENSOR = 0; + SELECTED_ROWS = 1; + } + required VarType type = 2; + optional LoDTensorDesc lod_desc = 3; + optional TensorDesc selected_rows_desc = 4; + optional bool persistable = 5 [ default = false ]; } ``` -## Definition of LodTensorDesc +## Definition of TensorDesc ```proto enum DataType { @@ -38,87 +45,25 @@ enum DataType { FP64 = 6; } -message LoDTensorDesc { +message TensorDesc { required DataType data_type = 1; - repeated int32 dims = 2; // [UNK, 640, 480] is saved as [-1, 640, 480] - optional int32 lod_level = 3 [default=0]; + repeated int64 dims = 2; // [UNK, 640, 480] is saved as [-1, 640, 480] } ``` -## Definition of Variable in Python - -In Python API, layer will take Variable as Input, and return Variable as Output. There should be a class `Variable` in python to help create and manage Variable. - -```python -image = Variable(dims=[-1, 640, 480]) -# fc1 and fc2 are both Variable -fc1 = layer.fc(input=image, output_size=10) -fc2 = layer.fc(input=fc1, output_size=20) -``` -### what should class `Variable` Have -1. `name`.a name of string type is used to mark the value of the Variable. -1. `initializer`. Since our Tensor does not have value. we will always use some Operator to fullfill it when run. So we should have a initialize method to help add the init operator. -1. `operator`. Variable should record which operator produce itself. The reaon is: - - we use pd.eval(targets=[var1, var2]) to run the related ops to get the value of var1 and var2. var.op is used to trace the dependency of the current variable. - -In PaddlePaddle, we use Block to describe Computation Graph, so in the code we will use Block but not Graph. - -```python -import VarDesc -import LoDTensorDesc -import framework - -def AddInitialOperator(variable, initializer): - # add an initialize Operator to block to init this Variable - -class Variable(object): - def __init__(self, name, dims, type, initializer): - self._block = get_default_block() - self._name = name - self.op = None - - tensor_desc = LoDTensorDesc(data_type=type, dims=dims) - _var_desc = VarDesc(name=name, lod_tensor=tensor_desc) - self._var = framework.CreateVar(_var_desc) - self._block.add_var(self) +A TensorDesc describes `SelectedRows` and `LoDTensor`. For details of `SelectedRows`, please reference [`SelectedRows`](./selected_rows.md). - # add initial op according to initializer - if initializer is not None: - AddInitialOperator(self, initializer) - - def dims(self): - return self._var.dims() - - def data_type(self): - return self._var.data_type() +## Definition of LodTensorDesc - def to_proto(self): - pass +```proto +message LoDTensorDesc { + required TensorDesc tensor = 1; + optional int lod_level = 2; +} ``` -Then we can use this Variable to create a fc layer in Python. +A LoDTensorDesc contains a tensor and a lod_level. -```python -import paddle as pd - -def flatten_size(X, num_flatten_dims): - prod = 1 # of last num_flatten_dims - for i in xrange(num_flatten_dims): - prod = prod * X.dims[-i-1] - return prod - -def layer.fc(X, output_size, num_flatten_dims): - W = Variable(pd.random_uniform(), type=FP32, dims=[flatten_size(X, num_flatten_dims), output_size]) - b = Variable(pd.random_uniform(), type=FP32, dims=[output_size]) - out = Variable(type=FP32) - y = operator.fc(X, W, b, output=out) # fc will put fc op input into out - pd.InferShape(y) - return out - -x = Variable(dims=[-1, 640, 480]) -y = layer.fc(x, output_size=100) -z = layer.fc(y, output_size=200) +## Definition of Variable in Python -paddle.eval(targets=[z], ...) -print(z) -``` +For Variable in Python, please reference [`Python API`](./python_api.md). diff --git a/doc/faq/build_and_install/index_cn.rst b/doc/faq/build_and_install/index_cn.rst new file mode 100644 index 0000000000000000000000000000000000000000..f1677e216f31d79b53ac29a0afbf6fbb886a0dcd --- /dev/null +++ b/doc/faq/build_and_install/index_cn.rst @@ -0,0 +1,111 @@ +################### +编译安装与单元测试 +################### + +.. contents:: + +1. 运行Docker GPU镜像出现 "CUDA driver version is insufficient" +---------------------------------------------------------------- + +用户在使用PaddlePaddle GPU的Docker镜像的时候，常常出现 `Cuda Error: CUDA driver version is insufficient for CUDA runtime version`, 原因在于没有把机器上CUDA相关的驱动和库映射到容器内部。 +具体的解决方法是： + +.. code-block:: bash + + $ export CUDA_SO="$(\ls usr/lib64/libcuda* | xargs -I{} echo '-v {}:{}') $(\ls /usr/lib64/libnvidia* | xargs -I{} echo '-v {}:{}')" + $ export DEVICES=$(\ls /dev/nvidia* | xargs -I{} echo '--device {}:{}') + $ docker run ${CUDA_SO} ${DEVICES} -it paddledev/paddlepaddle:latest-gpu + +更多关于Docker的安装与使用, 请参考 `PaddlePaddle Docker 文档 `_ 。 + + +2. CMake源码编译, 找到的PythonLibs和PythonInterp版本不一致 +---------------------------------------------------------------- + +这是目前CMake寻找Python的逻辑存在缺陷，如果系统安装了多个Python版本，CMake找到的Python库和Python解释器版本可能有不一致现象，导致编译PaddlePaddle失败。正确的解决方法是， +用户强制指定特定的Python版本，具体操作如下： + + .. code-block:: bash + + cmake .. -DPYTHON_EXECUTABLE= -DPYTHON_LIBRARY= -DPYTHON_INCLUDE_DIR= + +用户需要指定本机上Python的路径：````, ````, ```` + +3. CMake源码编译，Paddle版本号为0.0.0 +-------------------------------------- + +如果运行 :code:`paddle version`, 出现 :code:`PaddlePaddle 0.0.0`；或者运行 :code:`cmake ..`，出现 + +.. code-block:: bash + + CMake Warning at cmake/version.cmake:20 (message): + Cannot add paddle version from git tag + +那么用户需要拉取所有的远程分支到本机，命令为 :code:`git fetch upstream`，然后重新cmake即可。 + +4. paddlepaddle\*.whl is not a supported wheel on this platform. +------------------------------------------------------------------------ + +出现这个问题的主要原因是，没有找到和当前系统匹配的paddlepaddle安装包。最新的paddlepaddle python安装包支持Linux x86_64和MacOS 10.12操作系统，并安装了python 2.7和pip 9.0.1。 + +更新 :code:`pip` 包的方法是\: + +.. code-block:: bash + + pip install --upgrade pip + +如果还不行，可以执行 :code:`python -c "import pip; print(pip.pep425tags.get_supported())"` 获取当前系统支持的python包的后缀， +并对比是否和正在安装的后缀一致。 + +如果系统支持的是 :code:`linux_x86_64` 而安装包是 :code:`manylinux1_x86_64` ，需要升级pip版本到最新； +如果系统支持 :code:`manylinux1_x86_64` 而安装包（本地）是 :code:`linux_x86_64` ，可以重命名这个whl包为 :code:`manylinux1_x86_64` 再安装。 + +5. 编译安装后执行 import paddle.v2 as paddle 报ImportError: No module named v2 +------------------------------------------------------------------------------------------ +先查看一下是否曾经安装过paddle v1版本，有的话需要先卸载： + +pip uninstall py_paddle paddle + +然后安装paddle的python环境, 在build目录下执行 + +pip install python/dist/paddle*.whl && pip install ../paddle/dist/py_paddle*.whl + +6. 遇到“非法指令”或者是“illegal instruction” +-------------------------------------------- + +PaddlePaddle使用avx SIMD指令提高cpu执行效率，因此错误的使用二进制发行版可能会导致这种错误，请选择正确的版本。 + +7. python相关的单元测试都过不了 +-------------------------------- + +如果出现以下python相关的单元测试都过不了的情况： + +.. code-block:: bash + + 24 - test_PyDataProvider (Failed) + 26 - test_RecurrentGradientMachine (Failed) + 27 - test_NetworkCompare (Failed) + 28 - test_PyDataProvider2 (Failed) + 32 - test_Prediction (Failed) + 33 - test_Compare (Failed) + 34 - test_Trainer (Failed) + 35 - test_TrainerOnePass (Failed) + 36 - test_CompareTwoNets (Failed) + 37 - test_CompareTwoOpts (Failed) + 38 - test_CompareSparse (Failed) + 39 - test_recurrent_machine_generation (Failed) + 40 - test_PyDataProviderWrapper (Failed) + 41 - test_config_parser (Failed) + 42 - test_swig_api (Failed) + 43 - layers_test (Failed) + +并且查询PaddlePaddle单元测试的日志，提示： + +.. code-block:: bash + + paddle package is already in your PYTHONPATH. But unittest need a clean environment. + Please uninstall paddle package before start unittest. Try to 'pip uninstall paddle'. + +解决办法是： + +* 卸载PaddlePaddle包 :code:`pip uninstall paddle`, 清理掉老旧的PaddlePaddle安装包，使得单元测试有一个干净的环境。如果PaddlePaddle包已经在python的site-packages里面，单元测试会引用site-packages里面的python包，而不是源码目录里 :code:`/python` 目录下的python包。同时，即便设置 :code:`PYTHONPATH` 到 :code:`/python` 也没用，因为python的搜索路径是优先已经安装的python包。 diff --git a/doc/faq/cluster/index_cn.rst b/doc/faq/cluster/index_cn.rst new file mode 100644 index 0000000000000000000000000000000000000000..e59c1e1a54a0c876d1e6e89f88030de59fb9fc1a --- /dev/null +++ b/doc/faq/cluster/index_cn.rst @@ -0,0 +1,17 @@ +############### +集群训练与预测 +############### + +.. contents:: + +1. 集群多节点训练，日志中保存均为网络通信类错误 +------------------------------------------------ + +集群多节点训练，日志报错为网络通信类错误，比如 :code:`Connection reset by peer` 等。 +此类报错通常是由于某一个节点的错误导致这个节点的训练进程退出，从而引发其他节点无法连接导致，可以参考下面的步骤排查： + +* 从 :code:`train.log` ， :code:`server.log` 找到最早报错的地方，查看是否是其他错误引发的报错（比如FPE，内存不足，磁盘空间不足等）。 + +* 如果发现最早的报错就是网络通信的问题，很有可能是非独占方式执行导致的端口冲突，可以联系OP，看当前MPI集群是否支持resource=full参数提交，如果支持增加此参数提交，并更换job 端口。 + +* 如果当前MPI集群并不支持任务独占模式，可以联系OP是否可以更换集群或升级当前集群。 diff --git a/doc/faq/index_cn.rst b/doc/faq/index_cn.rst index 00192aa69bd487787a8743d5589a365eacbd4ff3..9929767cac212237b3e2c3a547ba9a3c9d5f0979 100644 --- a/doc/faq/index_cn.rst +++ b/doc/faq/index_cn.rst @@ -1,375 +1,11 @@ -#################### FAQ -#################### +==== -.. contents:: +.. toctree:: + :maxdepth: 1 -1. 如何减少内存占用 ---------------------------------- - -神经网络的训练本身是一个非常消耗内存和显存的工作，经常会消耗数10GB的内存和数GB的显存。 -PaddlePaddle的内存占用主要分为如下几个方面\: - -* DataProvider缓冲池内存（只针对内存） -* 神经元激活内存（针对内存和显存） -* 参数内存 （针对内存和显存） -* 其他内存杂项 - -其中，其他内存杂项是指PaddlePaddle本身所用的一些内存，包括字符串分配，临时变量等等，暂不考虑在内。 - -减少DataProvider缓冲池内存 -++++++++++++++++++++++++++ - -PyDataProvider使用的是异步加载，同时在内存里直接随即选取数据来做Shuffle。即 - -.. graphviz:: - - digraph { - rankdir=LR; - 数据文件 -> 内存池 -> PaddlePaddle训练 - } - -所以，减小这个内存池即可减小内存占用，同时也可以加速开始训练前数据载入的过程。但是，这 -个内存池实际上决定了shuffle的粒度。所以，如果将这个内存池减小，又要保证数据是随机的， -那么最好将数据文件在每次读取之前做一次shuffle。可能的代码为 - -.. literalinclude:: src/reduce_min_pool_size.py - -这样做可以极大的减少内存占用，并且可能会加速训练过程，详细文档参考 :ref:`api_pydataprovider2` 。 - -神经元激活内存 -++++++++++++++ - -神经网络在训练的时候，会对每一个激活暂存一些数据，如神经元激活值等。 -在反向传递的时候，这些数据会被用来更新参数。这些数据使用的内存主要和两个参数有关系， -一是batch size，另一个是每条序列(Sequence)长度。所以，其实也是和每个mini-batch中包含 -的时间步信息成正比。 - -所以做法可以有两种： - -* 减小batch size。 即在网络配置中 :code:`settings(batch_size=1000)` 设置成一个小一些的值。但是batch size本身是神经网络的超参数，减小batch size可能会对训练结果产生影响。 -* 减小序列的长度，或者直接扔掉非常长的序列。比如，一个数据集大部分序列长度是100-200, - 但是突然有一个10000长的序列，就很容易导致内存超限，特别是在LSTM等RNN中。 - -参数内存 -++++++++ - -PaddlePaddle支持非常多的优化算法(Optimizer)，不同的优化算法需要使用不同大小的内存。 -例如使用 :code:`adadelta` 算法，则需要使用等于权重参数规模大约5倍的内存。举例，如果参数保存下来的模型目录 -文件为 :code:`100M`， 那么该优化算法至少需要 :code:`500M` 的内存。 - -可以考虑使用一些优化算法，例如 :code:`momentum`。 - -2. 如何加速PaddlePaddle的训练速度 ---------------------------------- - -加速PaddlePaddle训练可以考虑从以下几个方面\： - -* 减少数据载入的耗时 -* 加速训练速度 -* 利用分布式训练驾驭更多的计算资源 - -减少数据载入的耗时 -++++++++++++++++++ - -使用\ :code:`pydataprovider`\ 时，可以减少缓存池的大小，同时设置内存缓存功能，即可以极大的加速数据载入流程。 -:code:`DataProvider` 缓存池的减小，和之前减小通过减小缓存池来减小内存占用的原理一致。 - -.. literalinclude:: src/reduce_min_pool_size.py - -同时 :code:`@provider` 接口有一个 :code:`cache` 参数来控制缓存方法，将其设置成 :code:`CacheType.CACHE_PASS_IN_MEM` 的话，会将第一个 :code:`pass` (过完所有训练数据即为一个pass)生成的数据缓存在内存里，在之后的 :code:`pass` 中，不会再从 :code:`python` 端读取数据，而是直接从内存的缓存里读取数据。这也会极大减少数据读入的耗时。 - - -加速训练速度 -++++++++++++ - -PaddlePaddle支持Sparse的训练，sparse训练需要训练特征是 :code:`sparse_binary_vector` 、 :code:`sparse_vector` 、或者 :code:`integer_value` 的任一一种。同时，与这个训练数据交互的Layer，需要将其Parameter设置成 sparse 更新模式，即设置 :code:`sparse_update=True` - -这里使用简单的 :code:`word2vec` 训练语言模型距离，具体使用方法为\: - -使用一个词前两个词和后两个词，来预测这个中间的词。这个任务的DataProvider为\: - -.. literalinclude:: src/word2vec_dataprovider.py - -这个任务的配置为\: - -.. literalinclude:: src/word2vec_config.py - - -利用更多的计算资源 -++++++++++++++++++ - -利用更多的计算资源可以分为一下几个方式来进行\: - -* 单机CPU训练 - - * 使用多线程训练。设置命令行参数 :code:`trainer_count`。 - -* 单机GPU训练 - - * 使用显卡训练。设置命令行参数 :code:`use_gpu`。 - * 使用多块显卡训练。设置命令行参数 :code:`use_gpu` 和 :code:`trainer_count` 。 - -* 多机训练 - - * 请参考 :ref:`cluster_train` 。 - - -3. 遇到“非法指令”或者是“illegal instruction” --------------------------------------------- - -PaddlePaddle使用avx SIMD指令提高cpu执行效率，因此错误的使用二进制发行版可能会导致这种错误，请选择正确的版本。 - -4. 如何选择SGD算法的学习率 --------------------------- - -在采用sgd/async_sgd进行训练时，一个重要的问题是选择正确的learning_rate。如果learning_rate太大，那么训练有可能不收敛，如果learning_rate太小，那么收敛可能很慢，导致训练时间过长。 - -通常做法是从一个比较大的learning_rate开始试，如果不收敛，那减少学习率10倍继续试验，直到训练收敛为止。那么如何判断训练不收敛呢？可以估计出如果模型采用不变的输出最小的cost0是多少。 - -如果训练过程的的cost明显高于这个常数输出的cost，那么我们可以判断为训练不收敛。举一个例子，假如我们是三分类问题，采用multi-class-cross-entropy作为cost，数据中0,1,2三类的比例为 :code:`0.2, 0.5, 0.3` , 那么常数输出所能达到的最小cost是 :code:`-(0.2*log(0.2)+0.5*log(0.5)+0.3*log(0.3))=1.03` 。如果训练一个pass（或者更早）后，cost还大于这个数，那么可以认为训练不收敛，应该降低学习率。 - - -5. 如何初始化参数 ------------------ - -默认情况下，PaddlePaddle使用均值0，标准差为 :math:`\frac{1}{\sqrt{d}}` 来初始化参数。其中 :math:`d` 为参数矩阵的宽度。这种初始化方式在一般情况下不会产生很差的结果。如果用户想要自定义初始化方式，PaddlePaddle目前提供两种参数初始化的方式\: - -* 高斯分布。将 :code:`param_attr` 设置成 :code:`param_attr=ParamAttr(initial_mean=0.0, initial_std=1.0)` -* 均匀分布。将 :code:`param_attr` 设置成 :code:`param_attr=ParamAttr(initial_max=1.0, initial_min=-1.0)` - -比如设置一个全连接层的参数初始化方式和bias初始化方式，可以使用如下代码。 - -.. code-block:: python - - hidden = fc_layer(input=ipt, param_attr=ParamAttr(initial_max=1.0, initial_min=-1.0), - bias_attr=ParamAttr(initial_mean=1.0, initial_std=0.0)) - -上述代码将bias全部初始化为1.0, 同时将参数初始化为 :code:`[1.0, -1.0]` 的均匀分布。 - -6. 如何共享参数 ---------------- - -PaddlePaddle的参数使用名字 :code:`name` 作为参数的ID，相同名字的参数，会共享参数。设置参数的名字，可以使用 :code:`ParamAttr(name="YOUR_PARAM_NAME")` 来设置。更方便的设置方式，是使得要共享的参数使用同样的 :code:`ParamAttr` 对象。 - -简单的全连接网络，参数共享的配置示例为\: - -.. literalinclude:: ../../python/paddle/trainer_config_helpers/tests/configs/shared_fc.py - -这里 :code:`hidden_a` 和 :code:`hidden_b` 使用了同样的parameter和bias。并且softmax层的两个输入也使用了同样的参数 :code:`softmax_param`。 - -7. \*-cp27mu-linux_x86_64.whl is not a supported wheel on this platform. ------------------------------------------------------------------------- - -出现这个问题的主要原因是，系统编译wheel包的时候，使用的 :code:`wheel` 包是最新的， -而系统中的 :code:`pip` 包比较老。具体的解决方法是，更新 :code:`pip` 包并重新编译PaddlePaddle。 -更新 :code:`pip` 包的方法是\: - -.. code-block:: bash - - pip install --upgrade pip - -8. python相关的单元测试都过不了 --------------------------------- - -如果出现以下python相关的单元测试都过不了的情况： - -.. code-block:: bash - - 24 - test_PyDataProvider (Failed) - 26 - test_RecurrentGradientMachine (Failed) - 27 - test_NetworkCompare (Failed) - 28 - test_PyDataProvider2 (Failed) - 32 - test_Prediction (Failed) - 33 - test_Compare (Failed) - 34 - test_Trainer (Failed) - 35 - test_TrainerOnePass (Failed) - 36 - test_CompareTwoNets (Failed) - 37 - test_CompareTwoOpts (Failed) - 38 - test_CompareSparse (Failed) - 39 - test_recurrent_machine_generation (Failed) - 40 - test_PyDataProviderWrapper (Failed) - 41 - test_config_parser (Failed) - 42 - test_swig_api (Failed) - 43 - layers_test (Failed) - -并且查询PaddlePaddle单元测试的日志，提示： - -.. code-block:: bash - - paddle package is already in your PYTHONPATH. But unittest need a clean environment. - Please uninstall paddle package before start unittest. Try to 'pip uninstall paddle'. - -解决办法是： - -* 卸载PaddlePaddle包 :code:`pip uninstall paddle`, 清理掉老旧的PaddlePaddle安装包，使得单元测试有一个干净的环境。如果PaddlePaddle包已经在python的site-packages里面，单元测试会引用site-packages里面的python包，而不是源码目录里 :code:`/python` 目录下的python包。同时，即便设置 :code:`PYTHONPATH` 到 :code:`/python` 也没用，因为python的搜索路径是优先已经安装的python包。 - - -9. 运行Docker GPU镜像出现 "CUDA driver version is insufficient" ----------------------------------------------------------------- - -用户在使用PaddlePaddle GPU的Docker镜像的时候，常常出现 `Cuda Error: CUDA driver version is insufficient for CUDA runtime version`, 原因在于没有把机器上CUDA相关的驱动和库映射到容器内部。 -具体的解决方法是： - -.. code-block:: bash - - $ export CUDA_SO="$(\ls usr/lib64/libcuda* | xargs -I{} echo '-v {}:{}') $(\ls /usr/lib64/libnvidia* | xargs -I{} echo '-v {}:{}')" - $ export DEVICES=$(\ls /dev/nvidia* | xargs -I{} echo '--device {}:{}') - $ docker run ${CUDA_SO} ${DEVICES} -it paddledev/paddlepaddle:latest-gpu - -更多关于Docker的安装与使用, 请参考 `PaddlePaddle Docker 文档 `_ 。 - - -10. CMake源码编译, 找到的PythonLibs和PythonInterp版本不一致 ----------------------------------------------------------------- - -这是目前CMake寻找Python的逻辑存在缺陷，如果系统安装了多个Python版本，CMake找到的Python库和Python解释器版本可能有不一致现象，导致编译PaddlePaddle失败。正确的解决方法是， -用户强制指定特定的Python版本，具体操作如下： - - .. code-block:: bash - - cmake .. -DPYTHON_EXECUTABLE= -DPYTHON_LIBRARY= -DPYTHON_INCLUDE_DIR= - -用户需要指定本机上Python的路径：````, ````, ```` - -11. CMake源码编译，Paddle版本号为0.0.0 --------------------------------------- - -如果运行 :code:`paddle version`, 出现 :code:`PaddlePaddle 0.0.0`；或者运行 :code:`cmake ..`，出现 - -.. code-block:: bash - - CMake Warning at cmake/version.cmake:20 (message): - Cannot add paddle version from git tag - -那么用户需要拉取所有的远程分支到本机，命令为 :code:`git fetch upstream`，然后重新cmake即可。 - -12. A protocol message was rejected because it was too big ----------------------------------------------------------- - -如果在训练NLP相关模型时，出现以下错误： - -.. code-block:: bash - - [libprotobuf ERROR google/protobuf/io/coded_stream.cc:171] A protocol message was rejected because it was too big (more than 67108864 bytes). To increase the limit (or to disable these warnings), see CodedInputStream::SetTotalBytesLimit() in google/protobuf/io/coded_stream.h. - F1205 14:59:50.295174 14703 TrainerConfigHelper.cpp:59] Check failed: m->conf.ParseFromString(configProtoStr) - -可能的原因是：传给dataprovider的某一个args过大，一般是由于直接传递大字典导致的。错误的define_py_data_sources2类似： - -.. code-block:: python - - src_dict = dict() - for line_count, line in enumerate(open(src_dict_path, "r")): - src_dict[line.strip()] = line_count - - define_py_data_sources2( - train_list, - test_list, - module="dataprovider", - obj="process", - args={"src_dict": src_dict}) - -解决方案是：将字典的地址作为args传给dataprovider，然后在dataprovider里面根据该地址加载字典。即define_py_data_sources2应改为： - -.. code-block:: python - - define_py_data_sources2( - train_list, - test_list, - module="dataprovider", - obj="process", - args={"src_dict_path": src_dict_path}) - -完整源码可参考 `seqToseq `_ 示例。 - -13. 如何指定GPU设备 -------------------- - -例如机器上有4块GPU，编号从0开始，指定使用2、3号GPU： - -* 方式1：通过 `CUDA_VISIBLE_DEVICES `_ 环境变量来指定特定的GPU。 - -.. code-block:: bash - - env CUDA_VISIBLE_DEVICES=2,3 paddle train --use_gpu=true --trainer_count=2 - -* 方式2：通过命令行参数 ``--gpu_id`` 指定。 - -.. code-block:: bash - - paddle train --use_gpu=true --trainer_count=2 --gpu_id=2 - - -14. 训练过程中出现 :code:`Floating point exception`, 训练因此退出怎么办? ------------------------------------------------------------------------- - -Paddle二进制在运行时捕获了浮点数异常，只要出现浮点数异常(即训练过程中出现NaN或者Inf)，立刻退出。浮点异常通常的原因是浮点数溢出、除零等问题。 -主要原因包括两个方面: - -* 训练过程中参数或者训练过程中的梯度尺度过大，导致参数累加，乘除等时候，导致了浮点数溢出。 -* 模型一直不收敛，发散到了一个数值特别大的地方。 -* 训练数据有问题，导致参数收敛到了一些奇异的情况。或者输入数据尺度过大，有些特征的取值达到数百万，这时进行矩阵乘法运算就可能导致浮点数溢出。 - -主要的解决办法是减小学习律或者对数据进行归一化处理。 - -15. 编译安装后执行 import paddle.v2 as paddle 报ImportError: No module named v2 ------------------------------------------------------------------------- -先查看一下是否曾经安装过paddle v1版本，有的话需要先卸载： - -pip uninstall py_paddle paddle - -然后安装paddle的python环境, 在build目录下执行 - -pip install python/dist/paddle*.whl && pip install ../paddle/dist/py_paddle*.whl - -16. PaddlePaddle存储的参数格式是什么，如何和明文进行相互转化 ---------------------------------------------------------- - -PaddlePaddle保存的模型参数文件内容由16字节头信息和网络参数两部分组成。头信息中，1~4字节表示PaddlePaddle版本信息，请直接填充0；5~8字节表示每个参数占用的字节数，当保存的网络参数为float类型时为4，double类型时为8；9~16字节表示保存的参数总个数。 - -将PaddlePaddle保存的模型参数还原回明文时，可以使用相应数据类型的 :code:`numpy.array` 加载具体网络参数，此时可以跳过PaddlePaddle模型参数文件的头信息。若在PaddlePaddle编译时，未指定按照double精度编译，默认情况下按照float精度计算，保存的参数也是float类型。这时在使用 :code:`numpy.array` 时，一般设置 :code:`dtype=float32` 。示例如下： - -.. code-block:: python - - def read_parameter(fname, width): - s = open(fname).read() - # skip header - vec = np.fromstring(s[16:], dtype=np.float32) - # width is the size of the corresponding layer - np.savetxt(fname + ".csv", vec.reshape(width, -1), - fmt="%.6f", delimiter=",") - - -将明文参数转化为PaddlePaddle可加载的模型参数时，首先构造头信息，再写入网络参数。下面的代码将随机生成的矩阵转化为可以被PaddlePaddle加载的模型参数。 - -.. code-block:: python - - def gen_rand_param(param_file, width, height, need_trans): - np.random.seed() - header = struct.pack("iil", 0, 4, height * width) - param = np.float32(np.random.rand(height, width)) - with open(param_file, "w") as fparam: - fparam.write(header + param.tostring()) - -17. 如何加载预训练参数 ------------------------------- - -* 对加载预训练参数的层，设置其参数属性 :code:`is_static=True`，使该层的参数在训练过程中保持不变。以embedding层为例，代码如下： - -.. code-block:: python - - emb_para = paddle.attr.Param(name='emb', is_static=True) - paddle.layer.embedding(size=word_dim, input=x, param_attr=emb_para) - - -* 从模型文件将预训练参数载入 :code:`numpy.array`，在创建parameters后，使用 :code:`parameters.set()` 加载预训练参数。PaddlePaddle保存的模型参数文件前16字节为头信息，用户将参数载入 :code:`numpy.array` 时须从第17字节开始。以embedding层为例，代码如下： - -.. code-block:: python - - def load_parameter(file_name, h, w): - with open(file_name, 'rb') as f: - f.read(16) # skip header. - return np.fromfile(f, dtype=np.float32).reshape(h, w) - - parameters = paddle.parameters.create(my_cost) - parameters.set('emb', load_parameter(emb_param_file, 30000, 256)) + build_and_install/index_cn.rst + model/index_cn.rst + parameter/index_cn.rst + local/index_cn.rst + cluster/index_cn.rst diff --git a/doc/faq/local/index_cn.rst b/doc/faq/local/index_cn.rst new file mode 100644 index 0000000000000000000000000000000000000000..0e939a2671ace8682c90cdc1c1bb2da1dda0d568 --- /dev/null +++ b/doc/faq/local/index_cn.rst @@ -0,0 +1,213 @@ +############### +本地训练与预测 +############### + +.. contents:: + +1. 如何减少内存占用 +------------------- + +神经网络的训练本身是一个非常消耗内存和显存的工作，经常会消耗数10GB的内存和数GB的显存。 +PaddlePaddle的内存占用主要分为如下几个方面\: + +* DataProvider缓冲池内存（只针对内存） +* 神经元激活内存（针对内存和显存） +* 参数内存 （针对内存和显存） +* 其他内存杂项 + +其中，其他内存杂项是指PaddlePaddle本身所用的一些内存，包括字符串分配，临时变量等等，暂不考虑在内。 + +减少DataProvider缓冲池内存 +++++++++++++++++++++++++++ + +PyDataProvider使用的是异步加载，同时在内存里直接随即选取数据来做Shuffle。即 + +.. graphviz:: + + digraph { + rankdir=LR; + 数据文件 -> 内存池 -> PaddlePaddle训练 + } + +所以，减小这个内存池即可减小内存占用，同时也可以加速开始训练前数据载入的过程。但是，这 +个内存池实际上决定了shuffle的粒度。所以，如果将这个内存池减小，又要保证数据是随机的， +那么最好将数据文件在每次读取之前做一次shuffle。可能的代码为 + +.. literalinclude:: src/reduce_min_pool_size.py + +这样做可以极大的减少内存占用，并且可能会加速训练过程，详细文档参考 :ref:`api_pydataprovider2` 。 + +神经元激活内存 +++++++++++++++ + +神经网络在训练的时候，会对每一个激活暂存一些数据，如神经元激活值等。 +在反向传递的时候，这些数据会被用来更新参数。这些数据使用的内存主要和两个参数有关系， +一是batch size，另一个是每条序列(Sequence)长度。所以，其实也是和每个mini-batch中包含 +的时间步信息成正比。 + +所以做法可以有两种： + +* 减小batch size。 即在网络配置中 :code:`settings(batch_size=1000)` 设置成一个小一些的值。但是batch size本身是神经网络的超参数，减小batch size可能会对训练结果产生影响。 +* 减小序列的长度，或者直接扔掉非常长的序列。比如，一个数据集大部分序列长度是100-200, + 但是突然有一个10000长的序列，就很容易导致内存超限，特别是在LSTM等RNN中。 + +参数内存 +++++++++ + +PaddlePaddle支持非常多的优化算法(Optimizer)，不同的优化算法需要使用不同大小的内存。 +例如使用 :code:`adadelta` 算法，则需要使用等于权重参数规模大约5倍的内存。举例，如果参数保存下来的模型目录 +文件为 :code:`100M`， 那么该优化算法至少需要 :code:`500M` 的内存。 + +可以考虑使用一些优化算法，例如 :code:`momentum`。 + +2. 如何加速训练速度 +------------------- + +加速PaddlePaddle训练可以考虑从以下几个方面\： + +* 减少数据载入的耗时 +* 加速训练速度 +* 利用分布式训练驾驭更多的计算资源 + +减少数据载入的耗时 +++++++++++++++++++ + +使用\ :code:`pydataprovider`\ 时，可以减少缓存池的大小，同时设置内存缓存功能，即可以极大的加速数据载入流程。 +:code:`DataProvider` 缓存池的减小，和之前减小通过减小缓存池来减小内存占用的原理一致。 + +.. literalinclude:: src/reduce_min_pool_size.py + +同时 :code:`@provider` 接口有一个 :code:`cache` 参数来控制缓存方法，将其设置成 :code:`CacheType.CACHE_PASS_IN_MEM` 的话，会将第一个 :code:`pass` (过完所有训练数据即为一个pass)生成的数据缓存在内存里，在之后的 :code:`pass` 中，不会再从 :code:`python` 端读取数据，而是直接从内存的缓存里读取数据。这也会极大减少数据读入的耗时。 + + +加速训练速度 +++++++++++++ + +PaddlePaddle支持Sparse的训练，sparse训练需要训练特征是 :code:`sparse_binary_vector` 、 :code:`sparse_vector` 、或者 :code:`integer_value` 的任一一种。同时，与这个训练数据交互的Layer，需要将其Parameter设置成 sparse 更新模式，即设置 :code:`sparse_update=True` + +这里使用简单的 :code:`word2vec` 训练语言模型距离，具体使用方法为\: + +使用一个词前两个词和后两个词，来预测这个中间的词。这个任务的DataProvider为\: + +.. literalinclude:: src/word2vec_dataprovider.py + +这个任务的配置为\: + +.. literalinclude:: src/word2vec_config.py + + +利用更多的计算资源 +++++++++++++++++++ + +利用更多的计算资源可以分为一下几个方式来进行\: + +* 单机CPU训练 + + * 使用多线程训练。设置命令行参数 :code:`trainer_count`。 + +* 单机GPU训练 + + * 使用显卡训练。设置命令行参数 :code:`use_gpu`。 + * 使用多块显卡训练。设置命令行参数 :code:`use_gpu` 和 :code:`trainer_count` 。 + +* 多机训练 + + * 请参考 :ref:`cluster_train` 。 + +3. 如何指定GPU设备 +------------------ + +例如机器上有4块GPU，编号从0开始，指定使用2、3号GPU： + +* 方式1：通过 `CUDA_VISIBLE_DEVICES `_ 环境变量来指定特定的GPU。 + +.. code-block:: bash + + env CUDA_VISIBLE_DEVICES=2,3 paddle train --use_gpu=true --trainer_count=2 + +* 方式2：通过命令行参数 ``--gpu_id`` 指定。 + +.. code-block:: bash + + paddle train --use_gpu=true --trainer_count=2 --gpu_id=2 + + +4. 训练过程中出现 :code:`Floating point exception`, 训练因此退出怎么办? +------------------------------------------------------------------------ + +Paddle二进制在运行时捕获了浮点数异常，只要出现浮点数异常(即训练过程中出现NaN或者Inf)，立刻退出。浮点异常通常的原因是浮点数溢出、除零等问题。 +主要原因包括两个方面: + +* 训练过程中参数或者训练过程中的梯度尺度过大，导致参数累加，乘除等时候，导致了浮点数溢出。 +* 模型一直不收敛，发散到了一个数值特别大的地方。 +* 训练数据有问题，导致参数收敛到了一些奇异的情况。或者输入数据尺度过大，有些特征的取值达到数百万，这时进行矩阵乘法运算就可能导致浮点数溢出。 + +这里有两种有效的解决方法： + +1. 设置 :code:`gradient_clipping_threshold` 参数，示例代码如下： + +.. code-block:: python + +optimizer = paddle.optimizer.RMSProp( + learning_rate=1e-3, + gradient_clipping_threshold=10.0, + regularization=paddle.optimizer.L2Regularization(rate=8e-4)) + +具体可以参考 `nmt_without_attention `_ 示例。 + +2. 设置 :code:`error_clipping_threshold` 参数，示例代码如下： + +.. code-block:: python + +decoder_inputs = paddle.layer.fc( + act=paddle.activation.Linear(), + size=decoder_size * 3, + bias_attr=False, + input=[context, current_word], + layer_attr=paddle.attr.ExtraLayerAttribute( + error_clipping_threshold=100.0)) + +完整代码可以参考示例 `machine translation `_ 。 + +两种方法的区别： + +1. 两者都是对梯度的截断，但截断时机不同，前者在 :code:`optimzier` 更新网络参数时应用；后者在激活函数反向计算时被调用； +2. 截断对象不同：前者截断可学习参数的梯度，后者截断回传给前层的梯度; + +除此之外，还可以通过减小学习率或者对数据进行归一化处理来解决这类问题。 + +5. 如何调用 infer 接口输出多个layer的预测结果 +----------------------------------------------- + +* 将需要输出的层作为 :code:`paddle.inference.Inference()` 接口的 :code:`output_layer` 参数输入，代码如下： + +.. code-block:: python + + inferer = paddle.inference.Inference(output_layer=[layer1, layer2], parameters=parameters) + +* 指定要输出的字段进行输出。以输出 :code:`value` 字段为例，代码如下： + +.. code-block:: python + + out = inferer.infer(input=data_batch, field=["value"]) + +需要注意的是： + +* 如果指定了2个layer作为输出层，实际上需要的输出结果是两个矩阵； +* 假设第一个layer的输出A是一个 N1 * M1 的矩阵，第二个 Layer 的输出B是一个 N2 * M2 的矩阵； +* paddle.v2 默认会将A和B 横向拼接，当N1 和 N2 大小不一样时，会报如下的错误： + +.. code-block:: python + + ValueError: all the input array dimensions except for the concatenation axis must match exactly + +多个层的输出矩阵的高度不一致导致拼接失败，这种情况常常发生在： + +* 同时输出序列层和非序列层； +* 多个输出层处理多个不同长度的序列; + +此时可以在调用infer接口时通过设置 :code:`flatten_result=False` , 跳过“拼接”步骤，来解决上面的问题。这时，infer接口的返回值是一个python list: + +* list 中元素的个数等于网络中输出层的个数； +* list 中每个元素是一个layer的输出结果矩阵，类型是numpy的ndarray； +* 每一个layer输出矩阵的高度，在非序列输入时：等于样本数；序列输入时等于：输入序列中元素的总数；宽度等于配置中layer的size； diff --git a/doc/faq/src/reduce_min_pool_size.py b/doc/faq/local/src/reduce_min_pool_size.py similarity index 100% rename from doc/faq/src/reduce_min_pool_size.py rename to doc/faq/local/src/reduce_min_pool_size.py diff --git a/doc/faq/src/word2vec_config.py b/doc/faq/local/src/word2vec_config.py similarity index 100% rename from doc/faq/src/word2vec_config.py rename to doc/faq/local/src/word2vec_config.py diff --git a/doc/faq/src/word2vec_dataprovider.py b/doc/faq/local/src/word2vec_dataprovider.py similarity index 100% rename from doc/faq/src/word2vec_dataprovider.py rename to doc/faq/local/src/word2vec_dataprovider.py diff --git a/doc/faq/model/index_cn.rst b/doc/faq/model/index_cn.rst new file mode 100644 index 0000000000000000000000000000000000000000..b47bbe05bdb39d1ade9434a7e54bf6ca88a91cc9 --- /dev/null +++ b/doc/faq/model/index_cn.rst @@ -0,0 +1,69 @@ +######### +模型配置 +######### + +.. contents:: + +1. 出现 :code:`Duplicated layer name` 错误怎么办 +-------------------------------------------------- + +出现该错误的原因一般是用户对不同layer的参数 :code:`name` 设置了相同的取值。遇到该错误时，先找出参数 :code:`name` 取值相同的layer，然后将这些layer的参数 :code:`name` 设置为不同的值。 + +2. :code:`paddle.layer.memory` 的参数 :code:`name` 如何使用 +------------------------------------------------------------- + +* :code:`paddle.layer.memory` 用于获取特定layer上一时间步的输出，该layer是通过参数 :code:`name` 指定，即，:code:`paddle.layer.memory` 会关联参数 :code:`name` 取值相同的layer，并将该layer上一时间步的输出作为自身当前时间步的输出。 + +* PaddlePaddle的所有layer都有唯一的name，用户通过参数 :code:`name` 设定，当用户没有显式设定时，PaddlePaddle会自动设定。而 :code:`paddle.layer.memory` 不是真正的layer，其name由参数 :code:`memory_name` 设定，当用户没有显式设定时，PaddlePaddle会自动设定。:code:`paddle.layer.memory` 的参数 :code:`name` 用于指定其要关联的layer，需要用户显式设定。 + +3. 两种使用 drop_out 的方法有何区别 +------------------------------------ + +* 在PaddlePaddle中使用dropout有两种方式 + + * 在相应layer的 :code:`layer_atter` 设置 :code:`drop_rate`，以 :code:`paddle.layer.fc` 为例，代码如下： + + .. code-block:: python + + fc = paddle.layer.fc(input=input, layer_attr=paddle.attr.ExtraLayerAttribute(drop_rate=0.5)) + + * 使用 :code:`paddle.layer.dropout`，以 :code:`paddle.layer.fc` 为例，代码如下： + + .. code-block:: python + + fc = paddle.layer.fc(input=input) + drop_fc = paddle.layer.dropout(input=fc, dropout_rate=0.5) + +* :code:`paddle.layer.dropout` 实际上使用了 :code:`paddle.layer.add_to`，并在该layer里采用第一种方式设置 :code:`drop_rate` 来使用dropout的。这种方式对内存消耗较大。 + +* PaddlePaddle在激活函数里实现dropout，而不是在layer里实现。 + +* :code:`paddle.layer.lstmemory`、:code:`paddle.layer.grumemory`、:code:`paddle.layer.recurrent` 不是通过一般的方式来实现对输出的激活，所以不能采用第一种方式在这几个layer里设置 :code:`drop_rate` 来使用dropout。若要对这几个layer使用dropout，可采用第二种方式，即使用 :code:`paddle.layer.dropout`。 + +4. 不同的 recurrent layer 的区别 +---------------------------------- +以LSTM为例，在PaddlePaddle中包含以下 recurrent layer： + +* :code:`paddle.layer.lstmemory` +* :code:`paddle.networks.simple_lstm` +* :code:`paddle.networks.lstmemory_group` +* :code:`paddle.networks.bidirectional_lstm` + +按照具体实现方式可以归纳为2类： + +1. 由 recurrent_group 实现的 recurrent layer： + + * 用户在使用这一类recurrent layer时，可以访问由recurrent unit在一个时间步内计算得到的中间值（例如：hidden states, memory cells等）； + * 上述的 :code:`paddle.networks.lstmemory_group` 是这一类的 recurrent layer ； + +2. 将recurrent layer作为一个整体来实现： + + * 用户在使用这一类recurrent layer，只能访问它们的输出值； + * 上述的 :code:`paddle.networks.lstmemory_group` 、 :code:`paddle.networks.simple_lstm` 和 :code:`paddle.networks.bidirectional_lstm` 属于这一类的实现； + +将recurrent layer作为一个整体来实现， 能够针对CPU和GPU的计算做更多优化， 所以相比于recurrent group的实现方式， 第二类 recurrent layer 计算效率更高。 在实际应用中，如果用户不需要访问LSTM的中间变量，而只需要获得recurrent layer计算的输出，我们建议使用第二类实现。 + +此外，关于LSTM, PaddlePaddle中还包含 :code:`paddle.networks.lstmemory_unit` 这一计算单元： + + * 不同于上述介绍的recurrent layer , :code:`paddle.networks.lstmemory_unit` 定义了LSTM单元在一个时间步内的计算过程，它并不是一个完整的recurrent layer，也不能接收序列数据作为输入； + * :code:`paddle.networks.lstmemory_unit` 只能在recurrent_group中作为step function使用； diff --git a/doc/faq/parameter/index_cn.rst b/doc/faq/parameter/index_cn.rst new file mode 100644 index 0000000000000000000000000000000000000000..6fa0c64413be1616a435640b0347904a49873349 --- /dev/null +++ b/doc/faq/parameter/index_cn.rst @@ -0,0 +1,201 @@ +######### +参数设置 +######### + +.. contents:: + +1. 如何选择SGD算法的学习率 +-------------------------- + +在采用sgd/async_sgd进行训练时，一个重要的问题是选择正确的learning_rate。如果learning_rate太大，那么训练有可能不收敛，如果learning_rate太小，那么收敛可能很慢，导致训练时间过长。 + +通常做法是从一个比较大的learning_rate开始试，如果不收敛，那减少学习率10倍继续试验，直到训练收敛为止。那么如何判断训练不收敛呢？可以估计出如果模型采用不变的输出最小的cost0是多少。 + +如果训练过程的的cost明显高于这个常数输出的cost，那么我们可以判断为训练不收敛。举一个例子，假如我们是三分类问题，采用multi-class-cross-entropy作为cost，数据中0,1,2三类的比例为 :code:`0.2, 0.5, 0.3` , 那么常数输出所能达到的最小cost是 :code:`-(0.2*log(0.2)+0.5*log(0.5)+0.3*log(0.3))=1.03` 。如果训练一个pass（或者更早）后，cost还大于这个数，那么可以认为训练不收敛，应该降低学习率。 + +2. 如何设置学习率退火（learning rate annealing） +------------------------------------------------ + +在相应的优化算法里设置learning_rate_schedule及相关参数，以使用Adam算法为例，代码如下： + +.. code-block:: python + + optimizer = paddle.optimizer.Adam( + learning_rate=1e-3, + learning_rate_decay_a=0.5, + learning_rate_decay_b=0.75, + learning_rate_schedule="poly",) + +PaddlePaddle目前支持8种learning_rate_schedule，这8种learning_rate_schedule及其对应学习率计算方式如下： + +* "constant" + + lr = learning_rate + +* "poly" + + lr = learning_rate * pow(1 + learning_rate_decay_a * num_samples_processed, -learning_rate_decay_b) + + 其中，num_samples_processed为已训练样本数，下同。 + +* "caffe_poly" + + lr = learning_rate * pow(1.0 - num_samples_processed / learning_rate_decay_a, learning_rate_decay_b) + +* "exp" + + lr = learning_rate * pow(learning_rate_decay_a, num_samples_processed / learning_rate_decay_b) + +* "discexp" + + lr = learning_rate * pow(learning_rate_decay_a, floor(num_samples_processed / learning_rate_decay_b)) + +* "linear" + + lr = max(learning_rate - learning_rate_decay_a * num_samples_processed, learning_rate_decay_b) + +* "manual" + + 这是一种按已训练样本数分段取值的学习率退火方法。使用该learning_rate_schedule时，用户通过参数 :code:`learning_rate_args` 设置学习率衰减因子分段函数，当前的学习率为所设置 :code:`learning_rate` 与当前的衰减因子的乘积。以使用Adam算法为例，代码如下： + + .. code-block:: python + + optimizer = paddle.optimizer.Adam( + learning_rate=1e-3, + learning_rate_schedule="manual", + learning_rate_args="1000:1.0,2000:0.9,3000:0.8",) + + 在该示例中，当已训练样本数小于等于1000时，学习率为 :code:`1e-3 * 1.0`；当已训练样本数大于1000小于等于2000时，学习率为 :code:`1e-3 * 0.9`；当已训练样本数大于2000时，学习率为 :code:`1e-3 * 0.8`。 + +* "pass_manual" + + 这是一种按已训练pass数分段取值的学习率退火方法。使用该learning_rate_schedule时，用户通过参数 :code:`learning_rate_args` 设置学习率衰减因子分段函数，当前的学习率为所设置 :code:`learning_rate` 与当前的衰减因子的乘积。以使用Adam算法为例，代码如下： + + .. code-block:: python + + optimizer = paddle.optimizer.Adam( + learning_rate=1e-3, + learning_rate_schedule="pass_manual", + learning_rate_args="1:1.0,2:0.9,3:0.8",) + + 在该示例中，当已训练pass数小于等于1时，学习率为 :code:`1e-3 * 1.0`；当已训练pass数大于1小于等于2时，学习率为 :code:`1e-3 * 0.9`；当已训练pass数大于2时，学习率为 :code:`1e-3 * 0.8`。 + +3. 如何初始化参数 +----------------- + +默认情况下，PaddlePaddle使用均值0，标准差为 :math:`\frac{1}{\sqrt{d}}` 来初始化参数。其中 :math:`d` 为参数矩阵的宽度。这种初始化方式在一般情况下不会产生很差的结果。如果用户想要自定义初始化方式，PaddlePaddle目前提供两种参数初始化的方式\: + +* 高斯分布。将 :code:`param_attr` 设置成 :code:`param_attr=ParamAttr(initial_mean=0.0, initial_std=1.0)` +* 均匀分布。将 :code:`param_attr` 设置成 :code:`param_attr=ParamAttr(initial_max=1.0, initial_min=-1.0)` + +比如设置一个全连接层的参数初始化方式和bias初始化方式，可以使用如下代码。 + +.. code-block:: python + + hidden = fc_layer(input=ipt, param_attr=ParamAttr(initial_max=1.0, initial_min=-1.0), + bias_attr=ParamAttr(initial_mean=1.0, initial_std=0.0)) + +上述代码将bias全部初始化为1.0, 同时将参数初始化为 :code:`[1.0, -1.0]` 的均匀分布。 + +4. 如何共享参数 +--------------- + +PaddlePaddle的参数使用名字 :code:`name` 作为参数的ID，相同名字的参数，会共享参数。设置参数的名字，可以使用 :code:`ParamAttr(name="YOUR_PARAM_NAME")` 来设置。更方便的设置方式，是使得要共享的参数使用同样的 :code:`ParamAttr` 对象。 + +简单的全连接网络，参数共享的配置示例为\: + +.. literalinclude:: ../../python/paddle/trainer_config_helpers/tests/configs/shared_fc.py + +这里 :code:`hidden_a` 和 :code:`hidden_b` 使用了同样的parameter和bias。并且softmax层的两个输入也使用了同样的参数 :code:`softmax_param`。 + +5. 如何加载预训练参数 +------------------------ + +* 对加载预训练参数的层，设置其参数属性 :code:`is_static=True`，使该层的参数在训练过程中保持不变。以embedding层为例，代码如下： + +.. code-block:: python + + emb_para = paddle.attr.Param(name='emb', is_static=True) + paddle.layer.embedding(size=word_dim, input=x, param_attr=emb_para) + + +* 从模型文件将预训练参数载入 :code:`numpy.array`，在创建parameters后，使用 :code:`parameters.set()` 加载预训练参数。PaddlePaddle保存的模型参数文件前16字节为头信息，用户将参数载入 :code:`numpy.array` 时须从第17字节开始。以embedding层为例，代码如下： + +.. code-block:: python + + def load_parameter(file_name, h, w): + with open(file_name, 'rb') as f: + f.read(16) # skip header. + return np.fromfile(f, dtype=np.float32).reshape(h, w) + + parameters = paddle.parameters.create(my_cost) + parameters.set('emb', load_parameter(emb_param_file, 30000, 256)) + +6. 存储的参数格式是什么，如何和明文进行相互转化 +-------------------------------------------------- + +PaddlePaddle保存的模型参数文件内容由16字节头信息和网络参数两部分组成。头信息中，1~4字节表示PaddlePaddle版本信息，请直接填充0；5~8字节表示每个参数占用的字节数，当保存的网络参数为float类型时为4，double类型时为8；9~16字节表示保存的参数总个数。 + +将PaddlePaddle保存的模型参数还原回明文时，可以使用相应数据类型的 :code:`numpy.array` 加载具体网络参数，此时可以跳过PaddlePaddle模型参数文件的头信息。若在PaddlePaddle编译时，未指定按照double精度编译，默认情况下按照float精度计算，保存的参数也是float类型。这时在使用 :code:`numpy.array` 时，一般设置 :code:`dtype=float32` 。示例如下： + +.. code-block:: python + + def read_parameter(fname, width): + s = open(fname).read() + # skip header + vec = np.fromstring(s[16:], dtype=np.float32) + # width is the size of the corresponding layer + np.savetxt(fname + ".csv", vec.reshape(width, -1), + fmt="%.6f", delimiter=",") + + +将明文参数转化为PaddlePaddle可加载的模型参数时，首先构造头信息，再写入网络参数。下面的代码将随机生成的矩阵转化为可以被PaddlePaddle加载的模型参数。 + +.. code-block:: python + + def gen_rand_param(param_file, width, height, need_trans): + np.random.seed() + header = struct.pack("iil", 0, 4, height * width) + param = np.float32(np.random.rand(height, width)) + with open(param_file, "w") as fparam: + fparam.write(header + param.tostring()) + +7. A protocol message was rejected because it was too big +------------------------------------------------------------ + +如果在训练NLP相关模型时，出现以下错误： + +.. code-block:: bash + + [libprotobuf ERROR google/protobuf/io/coded_stream.cc:171] A protocol message was rejected because it was too big (more than 67108864 bytes). To increase the limit (or to disable these warnings), see CodedInputStream::SetTotalBytesLimit() in google/protobuf/io/coded_stream.h. + F1205 14:59:50.295174 14703 TrainerConfigHelper.cpp:59] Check failed: m->conf.ParseFromString(configProtoStr) + +可能的原因是：传给dataprovider的某一个args过大，一般是由于直接传递大字典导致的。错误的define_py_data_sources2类似： + +.. code-block:: python + + src_dict = dict() + for line_count, line in enumerate(open(src_dict_path, "r")): + src_dict[line.strip()] = line_count + + define_py_data_sources2( + train_list, + test_list, + module="dataprovider", + obj="process", + args={"src_dict": src_dict}) + +解决方案是：将字典的地址作为args传给dataprovider，然后在dataprovider里面根据该地址加载字典。即define_py_data_sources2应改为： + +.. code-block:: python + + define_py_data_sources2( + train_list, + test_list, + module="dataprovider", + obj="process", + args={"src_dict_path": src_dict_path}) + +完整源码可参考 `seqToseq `_ 示例。 + + diff --git a/doc/getstarted/build_and_install/docker_install_cn.rst b/doc/getstarted/build_and_install/docker_install_cn.rst index 84e33177740ca1652efc09c8081c2519b4366906..0d34dec8e908c5e61001500725187a2233797f46 100644 --- a/doc/getstarted/build_and_install/docker_install_cn.rst +++ b/doc/getstarted/build_and_install/docker_install_cn.rst @@ -20,7 +20,7 @@ Docker使用入门 docker pull paddlepaddle/paddle:0.10.0 - 来下载Docker镜像，paddlepaddle/paddle是从官方镜像源Dockerhub.com下载的，推荐国内用户使用ocker.paddlepaddle.org/paddle下载。 + 来下载Docker镜像，paddlepaddle/paddle是从官方镜像源Dockerhub.com下载的，推荐国内用户使用docker.paddlepaddle.org/paddle下载。 - *容器*： 如果说一个Docker镜像就是一个程序，那容器就是这个程序运行时产生的“进程”。 实际上，一个容器就是一个操作系统的进程，但是是运行在独立的进程空间，文件系统以及网络之上。 @@ -145,7 +145,7 @@ PaddlePaddle发布新版本的时候都会发布对应版本的生产镜像以 Jupyter Notebook是一个开源的web程序，大家可以通过它制作和分享带有代码、公式、图表、文字的交互式文档。用户可以通过网页浏览文档。 -PaddlePaddle Book是为用户和开发者制作的一个交互式的Jupyter Nodebook。 +PaddlePaddle Book是为用户和开发者制作的一个交互式的Jupyter Notebook。 如果您想要更深入了解deep learning，PaddlePaddle Book一定是您最好的选择。 我们提供可以直接运行PaddlePaddle Book的Docker镜像，直接运行： diff --git a/doc/howto/cross_compiling/cross_compiling_for_android_cn.md b/doc/howto/cross_compiling/cross_compiling_for_android_cn.md deleted file mode 100644 index 90dc84718c9ce1374cda6022de177afeeb60279d..0000000000000000000000000000000000000000 --- a/doc/howto/cross_compiling/cross_compiling_for_android_cn.md +++ /dev/null @@ -1,75 +0,0 @@ -# 构建Android平台上的PaddlePaddle库 - -用户可通过交叉编译的方式，在用户熟悉的开发平台（Linux，Mac OS X和Windows）上编译Android平台上适用的PaddlePaddle库。 -本文档将以Linux x86-64平台为例，介绍交叉编译Android平台上适用的PaddlePaddle库的方法和步骤。 - -## 准备交叉编译环境 - -从源码交叉编译PaddlePaddle，用户需要提前准备好交叉编译环境。Android平台上使用的C/C++交叉编译工具链为[Android NDK](https://developer.android.com/ndk/downloads/index.html?hl=zh-cn)，用户可自行前往下载预编译好的版本，也可通过以下命令获取： - -```bash -wget -q https://dl.google.com/android/repository/android-ndk-r14b-linux-x86_64.zip -unzip -q android-ndk-r14b-linux-x86_64.zip -``` - -Android NDK中包含了所有Android API级别、所有架构（arm/arm64/x86/mips）需要用到的编译工具和系统库。用户可根据自己的编译目标架构、所需支持的最低Android API级别，构建[独立工具链](https://developer.android.google.cn/ndk/guides/standalone_toolchain.html?hl=zh-cn)。 -比如： - -```bash -your/path/to/android-ndk-r14b-linux-x86_64/build/tools/make-standalone-toolchain.sh \ - --arch=arm --platform=android-21 --install-dir=your/path/to/my_standalone_toolchain -``` - -此命令将在your/path/to/my_standalone_toolchain目录生成一套编译工具链，面向架构为32位ARM架构，支持的最小的Android API级别为21，使用的编译器为arm-linux-androideabi-gcc (GCC) 4.9。 - -注意：**PaddlePaddle要求使用的编译工具链所支持的Andoid API级别不小于21**。 - -## 配置交叉编译参数 - -CMake系统对交叉编译提供了支持[cmake-toolchains](https://cmake.org/cmake/help/v3.0/manual/cmake-toolchains.7.html#cross-compiling)。为了简化cmake配置，PaddlePaddle为交叉编译提供了工具链配置文档[cmake/cross_compiling/android.cmake](https://github.com/PaddlePaddle/Paddle/blob/develop/cmake/cross_compiling/android.cmake)，以提供一些默认的编译器和编译参数相关配置。注意，从CMake 3.7版本开始，CMake官方对Android平台的交叉编译提供了通用的支持。PaddlePaddle若检测到用户使用的CMake版本不低于3.7时，将会将用户传进来的配置参数传递CMake系统，交由CMake系统本身来处理。有关参数配置的详细说明见[cmake-toolchains](https://cmake.org/cmake/help/v3.7/manual/cmake-toolchains.7.html#cross-compiling)。 - -交叉编译Android版本的PaddlePaddle库时，有一些必须配置的参数： -- `CMAKE_SYSTEM_NAME`，CMake编译的目标平台，必须设置为`Android`。在设置`CMAKE_SYSTEM_NAME=Android`后，PaddlePaddle的CMake系统才认为是在交叉编译Android系统的版本，并自动编译宿主机版protoc可执行文件、目标机版protobuf库、以及Android所需`arm_soft_fp_abi`分支的目标机版OpenBLAS库。此外，还会强制设置一些PaddlePaddle参数的值（`WITH_GPU=OFF`、`WITH_AVX=OFF`、`WITH_PYTHON=OFF`、`WITH_RDMA=OFF`）。 -- `WITH_C_API`，必须设置为`ON`。在Android平台上只支持使用C-API来预测。 -- `WITH_SWIG_PY`，必须设置为`OFF`。在Android平台上不支持通过swig调用来训练或者预测。 - -Android平台可选配置参数： - -- `ANDROID_STANDALONE_TOOLCHAIN`，独立工具链所在的绝对路径，或者相对于构建目录的相对路径。PaddlePaddle的CMake系统将根据该值自动推导和设置需要使用的交叉编译器、sysroot、以及Android API级别；否则，用户需要在cmake时手动设置这些值。无默认值。 -- `ANDROID_ABI`，目标架构ABI。目前只支持`armeabi-v7a`，默认值为`armeabi-v7a`。 -- `ANDROID_NATIVE_API_LEVEL`，工具链的Android API级别。若没有显式设置，PaddlePaddle将根据`ANDROID_STANDALONE_TOOLCHAIN`的值自动推导得到。 -- `ANROID_ARM_MODE`，是否使用ARM模式。可设置`ON/OFF`，默认值为`ON`。 -- `ANDROID_ARM_NEON`，是否使用NEON指令。目前必须设置成`ON`，默认值为`ON`。 - -其他配置参数： - -- `HOST_C/CXX_COMPILER`，宿主机的C/C++编译器。在编译宿主机版protoc可执行文件和目标机版OpenBLAS库时需要用到。默认设置成环境变量`CC`的值；若环境变量`CC`没有设置，则设置成`cc`编译器。 - -一种常用的cmake配置如下： - -```bash -cmake -DCMAKE_SYSTEM_NAME=Android \ - -DANDROID_STANDALONE_TOOLCHAIN=your/path/to/my_standalone_toolchain \ - -DANDROID_ABI=armeabi-v7a \ - -DANDROID_ARM_NEON=ON \ - -DANDROID_ARM_MODE=ON \ - -DCMAKE_INSTALL_PREFIX=your/path/to/install \ - -DWITH_C_API=ON \ - -DWITH_SWIG_PY=OFF \ - .. -``` - -用户还可根据自己的需求设置其他编译参数。比如希望最小化生成的库的大小，可以设置`CMAKE_BUILD_TYPE`为`MinSizeRel`；若希望最快的执行速度，则可设置`CMAKE_BUILD_TYPE`为`Release`。亦可以通过手动设置`CMAKE_C/CXX_FLAGS_MINSIZEREL/RELEASE`来影响PaddlePaddle的编译过程。 - -## 编译和安装 - -CMake配置完成后，执行以下命令，PaddlePaddle将自动下载和编译所有第三方依赖库、编译和安装PaddlePaddle预测库。 - -```bash -make -make install -``` - -注意：如果你曾经在源码目录下编译过其他平台的PaddlePaddle库，请先使用`rm -rf`命令删除`third_party`目录和`build`目录，以确保所有的第三方依赖库和PaddlePaddle代码都是针对新的CMake配置重新编译的。 - -执行完安装命令后，`your/path/to/install`目录中会包含`include`和`lib`目录，其中`include`中包含C-API的头文件，`lib`中包含一个Android版本的库。自此，PaddlePaddle的已经安装完成，用户可将`your/path/to/install`目录下的生成文件用于深度学习相关Android App中，调用方法见C-API文档。 diff --git a/doc/howto/cross_compiling/cross_compiling_for_raspberry_cn.md b/doc/howto/cross_compiling/cross_compiling_for_raspberry_cn.md deleted file mode 100644 index 085b5dda1615a9af918b59870db460fcc5acdcca..0000000000000000000000000000000000000000 --- a/doc/howto/cross_compiling/cross_compiling_for_raspberry_cn.md +++ /dev/null @@ -1,65 +0,0 @@ -# 构建Raspberry Pi平台上的PaddlePaddle库 - -对于Rasspberry Pi系统，用户可通过ssh等方式登录到Raspberry Pi系统上，按照[源码编译PaddlePaddle](http://www.paddlepaddle.org/doc_cn/getstarted/build_and_install/cmake/build_from_source_cn.html)相关文档所述，直接编译Raspberry Pi平台上适用的PaddlePaddle库。 - -用户也可以在自己熟悉的开发平台上，通过交叉编译的方式来编译。这篇文档将以Linux x86-64平台为例，介绍交叉编译Raspberry Pi平台上适用的PaddlePaddle的方法和步骤。 - -## 准备交叉编译环境 - -从源码交叉编译PaddlePaddle，用户需要提前准备好交叉编译环境。用户可自行前往[github](https://github.com/raspberrypi/tools)下载Raspberry Pi平台使用的C/C++交叉编译工具链，也可通过以下命令获取： - -```bash -git clone https://github.com/raspberrypi/tools.git -``` - -该github仓库中包含若干个预编译好的、针对不同平台的编译工具。宿主机是Linux x86-64环境，则需选用`arm-bcm2708/gcc-linaro-arm-linux-gnueabihf-raspbian-x64`下的作为编译工具，所使用的编译器为arm-linux-gnueabihf-gcc 4.8.3。 - -注意，该编译工具链需要系统glibc支持2.14以上。 - -## 配置交叉编译参数 - -CMake系统对交叉编译提供了支持[cmake-toolchains](https://cmake.org/cmake/help/v3.0/manual/cmake-toolchains.7.html#cross-compiling)。为了简化cmake配置，PaddlePaddle为交叉编译提供了工具链配置文档[cmake/cross_compiling/raspberry_pi.cmake](https://github.com/PaddlePaddle/Paddle/blob/develop/cmake/cross_compiling/raspberry_pi.cmake)，以提供一些默认的编译器和编译参数相关配置。 - -交叉编译Raspberry Pi版本PaddlePaddle库时，有一些必须配置的参数： - -- `CMAKE_SYSTEM_NAME`，CMake编译的目标平台，必须配置为`RPi`。在设置`CMAKE_SYSTEM_NAME=RPi`后，PaddlePaddle的CMake系统才认为在是在交叉编译Raspberry Pi系统的版本，并自动编译宿主机版protoc可执行文件、目标机版protobuf库、以及目标机版OpenBLAS库。 - -Raspberry Pi平台可选配置参数： - -- `RPI_TOOLCHAIN`，编译工具链所在的绝对路径，或者相对于构建目录的相对路径。PaddlePaddle的CMake系统将根据该值自动设置需要使用的交叉编译器；否则，用户需要在cmake时手动设置这些值。无默认值。 -- `RPI_ARM_NEON`，是否使用NEON指令。目前必须设置成`ON`，默认值为`ON`。 - -其他配置参数： - -- `HOST_C/CXX_COMPILER`，宿主机的C/C++编译器。在编译宿主机版protoc可执行文件和目标机版OpenBLAS库时需要用到。默认设置成环境变量`CC`的值；若环境变量`CC`没有设置，则设置成`cc`编译器。 - -cmake参数如下； - -``` -cmake -DCMAKE_SYSTEM_NAME=RPi \ - -DRPI_TOOLCHAIN=your/path/to/arm-bcm2708/gcc-linaro-arm-linux-gnueabihf-raspbian-x64 \ - -DRPI_ARM_NEON=ON \ - -DCMAKE_INSTALL_PREFIX=your/path/to/install \ - -DWITH_GPU=OFF \ - -DWITH_C_API=ON \ - -DWITH_PYTHON=OFF \ - -DWITH_SWIG_PY=OFF \ - .. -``` - -用户还可根据自己的需求设置其他编译参数。比如希望最小化生成的库的大小，可以设置`CMAKE_BUILD_TYPE`为`MinSizeRel`；若希望最快的执行速度，则可设置`CMAKE_BUILD_TYPE`为`Release`。亦可以通过手动设置`CMAKE_C/CXX_FLAGS_MINSIZEREL/RELEASE`来影响PaddlePaddle的编译过程。 - -## 编译和安装 - -CMake配置完成后，执行以下命令，PaddlePaddle将自动下载和编译所有第三方依赖库、编译和安装PaddlePaddle。 - -```bash -make -make install -``` - -注意：如果你曾经在源码目录下编译过其他平台的PaddlePaddle库，请先使用`rm -rf`命令删除`third_party`目录和`build`目录，以确保所有的第三方依赖库和PaddlePaddle代码都是针对新的CMake配置重新编译的。 - -执行完安装命令后，由于上一步cmake配置中`WITH_C_API`设置为`ON`，`your/path/to/install`目录中会包含`include`和`lib`目录，其中`include`中包含C-API的头文件，`lib`中包含一个Raspberry Pi版本的库。 - -更多的编译配置见[源码编译PaddlePaddle](http://www.paddlepaddle.org/doc_cn/getstarted/build_and_install/cmake/build_from_source_cn.html)相关文档。 diff --git a/doc/howto/deep_model/rnn/rnn_config_cn.rst b/doc/howto/deep_model/rnn/rnn_config_cn.rst index 4d684cf8ad5a8082cf31fb27027119b3d3e700b6..63fa161fafed0f3a8ec8799af21304cbec62d813 100644 --- a/doc/howto/deep_model/rnn/rnn_config_cn.rst +++ b/doc/howto/deep_model/rnn/rnn_config_cn.rst @@ -21,7 +21,7 @@ wmt14数据的提供文件在 `python/paddle/v2/dataset/wmt14.py