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/CMakeLists.txt b/CMakeLists.txt index 4783095194dc9c6409dc31c95588f46c9bee7c61..264420ad830ed39b38f1918951d8d66c84fd5ee9 100644 --- a/CMakeLists.txt +++ b/CMakeLists.txt @@ -105,6 +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 @@ -121,6 +127,7 @@ 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/nccl) include(cudnn) # set cudnn libraries, must before configure include(configure) # add paddle env configuration @@ -153,7 +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} ${NCCL_LIBRARY}) endif(NOT WITH_DSO) endif(WITH_GPU) diff --git a/Dockerfile b/Dockerfile index 136db772cc6a24b8084120fa6bab666bc1eda78e..150344a8116e2be9b5bab8e5fdcc9c37f4025020 100644 --- a/Dockerfile +++ b/Dockerfile @@ -22,7 +22,7 @@ COPY ./paddle/scripts/docker/root/ /root/ RUN apt-get update && \ apt-get install -y \ - git python-pip python-dev openssh-server bison \ + git python-pip python-dev openssh-server bison libnccl-dev \ wget unzip unrar tar xz-utils bzip2 gzip coreutils ntp \ curl sed grep graphviz libjpeg-dev zlib1g-dev \ python-matplotlib gcc-4.8 g++-4.8 \ diff --git a/cmake/configure.cmake b/cmake/configure.cmake index db8f5ab0456792f903093b9cf20e2541f00add5c..24ddb24399dabeec9b8e5faf36be3eb21f420111 100644 --- a/cmake/configure.cmake +++ b/cmake/configure.cmake @@ -62,11 +62,11 @@ else() FIND_PACKAGE(CUDA REQUIRED) if(${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/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..57d2c0a352507afd01d1cbf2c7b23c00ff7ad81b --- /dev/null +++ b/cmake/external/nccl.cmake @@ -0,0 +1,49 @@ +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/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/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=${WARPCTC_INSTALL_DIR} ) diff --git a/cmake/external/zlib.cmake b/cmake/external/zlib.cmake index c496a52b780364f3014f8fa3dfbc944a7aa7430e..e2c9fe56f335ae5b627b4d8d4bb17e4a2a466677 100644 --- a/cmake/external/zlib.cmake +++ b/cmake/external/zlib.cmake @@ -42,11 +42,11 @@ ExternalProject_Add( -DBUILD_SHARED_LIBS=OFF -DCMAKE_POSITION_INDEPENDENT_CODE=ON -DCMAKE_MACOSX_RPATH=ON - -DCMAKE_BUILD_TYPE=Release + -DCMAKE_BUILD_TYPE=${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 ff9868fc4e0d970b11e4763d2e0c8581f4f85907..c311783aa3187678c31c27ddbbd074790ca444f3 100644 --- a/cmake/generic.cmake +++ b/cmake/generic.cmake @@ -389,13 +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/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 9c812732d6ead76eb3aa2d1b617449c96807f21a..4066122c0e8dfa33776796c3d205ba5aec9e0f52 100644 --- a/doc/design/block.md +++ b/doc/design/block.md @@ -189,7 +189,7 @@ OpDesc { 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" : } }; @@ -243,7 +243,7 @@ class SymbolTable { // 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* NewVar(const string& name=""); + VarDesc* Var(const string& name=""); // find a VarDesc by name, if recursive is true, find parent's SymbolTable // recursively. 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/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/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/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/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 index 17440fae5028cfac5d58fc079ca2096d0be3a0f9..202b4b65103c0b7c536a9cb466c4120ce134d8c3 100644 --- a/doc/design/optimizer.md +++ b/doc/design/optimizer.md @@ -65,20 +65,6 @@ class Optimizer(object): def __init__(self): pass - def create_backward_pass(self, loss, parameter_list=None): - """ - create and add gradient Operators in BlockDesc to Compute gradients of `loss` - for parameters in parameter_list - - Args: - loss: an variable generated by cost function. - parameter_list: parameters that need to compute gradient and update to optimize the lost. - - Returns: - list of (parameters, gradients) pair. - """ - return None - def create_optimization_pass(self, parameters_and_grads): """Add optimization operators to update gradients to variables. @@ -93,7 +79,7 @@ class Optimizer(object): def minimize(self, loss, parameter_list): """Add operations to minimize `loss` by updating `parameter_list`. - This method combines interface `create_backward_pass()` and + This method combines interface `append_backward_ops()` and `create_optimization_pass()` into one. """ params_grads = self.create_backward_pass(loss, parameter_list) 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 index 56ae1d925a96622b5576013f38e33e5f92cbbb90..cb5fdc765b7126fc66a1c8978d4b96c0dc5a9f2c 100644 --- a/doc/design/python_api.md +++ b/doc/design/python_api.md @@ -179,40 +179,104 @@ init_attr={ `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 Functions +## Layer Function -A layer is a Python function that creates some operators and variables. Layers simplify the work of application programmers. +A layer is a Python function that creates some operators and variables. Layers simplify the work of application programmers. -### Data Layer +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 data_layer(name, type, column_name): - block = the_current_program.glolal_block() - var = block.create_global_var( - name=name, - shape=[None] + type.dims(), - dtype=type.dtype) - block.prepend_operator(block, - type="Feed", - inputs = None, - outputs = [var], - {column_name: column_name}) - return var +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 ``` -The input to the feed operator is a special variable in the global scope, which is the output of [Python readers](https://github.com/PaddlePaddle/Paddle/blob/develop/doc/design/reader/README.md). +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 -### FC Layer +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 -def fc_layer(input, size, ...): - block = program.current_block() - w = block.create_parameter(...) - b = block.create_parameter(...) - out = block.create_var() - op = block.append_operator("FC", X=input, W=w, b=b, out=out) - out.writer = op - return out +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 diff --git a/doc/design/refactorization.md b/doc/design/refactorization.md index ec51aa1a0ec667175ff7215dcd359023e296769f..f93d6155e1764386b01d2f0df3f141ab75cd55d4 100644 --- a/doc/design/refactorization.md +++ b/doc/design/refactorization.md @@ -177,9 +177,6 @@ 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. - --- # Registration Process 1. Write an Op class and its gradient Op class, if required. @@ -188,8 +185,6 @@ Make sure the registration process is executed and linked. 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` -4. Invoke the `USE` macro in which the Op is used to make sure that it is linked. - --- # Backward Module (1/2) ### Create Backward Operator diff --git a/doc/design/register_grad_op.md b/doc/design/register_grad_op.md index 3cf8a59446d244bb3a388b87b14273d9096c839a..8d973eb53178c3e889c845144553a453e11f067c 100644 --- a/doc/design/register_grad_op.md +++ b/doc/design/register_grad_op.md @@ -3,15 +3,17 @@ ## The Problem Posed -In our current operator registration mechanism, for each operator, the programmer should register a *gradient operator creator* function, which takes a C++ operator instance, and returns the corresponding gradient instance. +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, as we decided to separate the *compilation* and *execution* of DL models, we need to reshape the creator to take a protobuf `OpDesc` message, and returns a corresponding message. +However, we noticed two problems with the current design: -More than that, the new registration mechanism need to support the fact that an operators' gradient computation might be a composition of operators. +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. -## Current Implementation +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. -OpInfos store in a association map which key is the operator type. The `grad_op_type` indicate associated gradient operator type. Operator can create gradient operator by `OpInfo::creator_` of gradient. The pseudo code is +## 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 { @@ -29,16 +31,16 @@ OperatorBase* CreateGradientOperator(const OperatorBase& op) { ## Proposed Solution -The mapping relationship between an operator and its gradient operators is a function. The interface of that function is: +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 protobuf message `OpDesc` to manipulate `OpDesc` fast. +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`, to replace `grad_op_type_` field. The `OpInfo` should be +The `GradOpDescMaker` will be registered in `OpInfo` and will replace the `grad_op_type_` field. The `OpInfo` should look like ```cpp struct OpInfo { @@ -47,7 +49,7 @@ struct OpInfo { }; ``` -The `grad_op_maker_ ` is `nullptr` if the operator does not have associated gradient operators. +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 @@ -72,7 +74,7 @@ func = [] (const OpDescBind& fwd_op) { 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 chagne 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__`. +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 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 index 9e6f3b20cbcdc55e481fbe7bf5fa555d8b3c3d45..1a98839a957612b91b2276b58818623ecc62d1d5 100644 --- a/doc/design/selected_rows.md +++ b/doc/design/selected_rows.md @@ -1,6 +1,6 @@ # Design Doc: Selected Rows -`SelectedRows` is a kind 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 that tensor. It is straightforward to represent the sparse tensor by the following sparse tensor data structure: +`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 { @@ -11,7 +11,7 @@ class SelectedRows { }; ``` -The field `height_` shows the first dimension of `SelectedRows`. The `rows` are the indices of which rows of `SelectedRows` are non-zeros. The `value_` field is an N-dim tensor and shape is `[rows.size() /* NUM_ROWS */, ...]`, which supplies values for each row. The dimension of `SelectedRows` satisfies `[height_] + value_.shape[1:]`. +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: @@ -25,7 +25,7 @@ x = SelectedRow { ## SelectedRows in Protobuf -`SelectedRows` is a kind of `Variable`. `VarDesc` in protobuf should describe the `SelectedRows` information. Only the tensor dimension of a `SelectedRows` will be described in compile-time since the `rows_` and `value_` are related to training data. +`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 @@ -54,7 +54,7 @@ message VarDesc { ## InferShape for Selected Rows -Just like `LoD` information, `InferShape` method will inference output tensor type as well. The operator should decide whether its output is a `SelectedRows` or `Dense` tensor. +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 @@ -68,7 +68,7 @@ void TableLookupGrad::InferShape(context) { ## Sparse Operators -There are several operators should be written to support `SelectedRows`. They are: +There are several operators that need to be written to support `SelectedRows`. These are: -1. Operators which generates `SelectedRows` gradient. e.g. Gradient of `TableLookupOp`. +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 index 8378e97bf7cfaae54c36b1b92e202b16e4fe1e28..37e4f7b90f94fa3eb015e733999cd84c96b2239c 100644 --- a/doc/design/tensor_array.md +++ b/doc/design/tensor_array.md @@ -161,7 +161,7 @@ class TensorArray: @name: str the name of the variable to output. ''' - tensor = NewVar(name) + tensor = Var(name) tensor_array_stack(self.name, tensor) return tensor 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/local/index_cn.rst b/doc/faq/local/index_cn.rst index 75c4ba028e497e29e9030a86514348726d9c0a80..0e939a2671ace8682c90cdc1c1bb2da1dda0d568 100644 --- a/doc/faq/local/index_cn.rst +++ b/doc/faq/local/index_cn.rst @@ -174,7 +174,7 @@ decoder_inputs = paddle.layer.fc( 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律或者对数据进行归一化处理来解决这类问题。 +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 ----------------------------------------------- diff --git a/doc/howto/cross_compiling/cross_compiling_for_android_cn.md b/doc/howto/cross_compiling/cross_compiling_for_android_cn.md index 90dc84718c9ce1374cda6022de177afeeb60279d..1fc58c37cc9151d5e4d99b939e30c29aa99e04f1 100644 --- a/doc/howto/cross_compiling/cross_compiling_for_android_cn.md +++ b/doc/howto/cross_compiling/cross_compiling_for_android_cn.md @@ -1,9 +1,46 @@ # 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。 +用户可通过如下两种方式，交叉编译Android平台上适用的PaddlePaddle库： +- 基于Docker容器的编译方式 +- 基于Linux交叉编译环境的编译方式 + +## 基于Docker容器的编译方式 +Docker能在所有主要操作系统（包括Linux，Mac OS X和Windows）上运行，因此，使用基于Docker容器的编译方式，用户可在自己熟悉的开发平台上编译Android平台上适用的PaddlePaddle库。 + +### 构建PaddlePaddle的Android开发镜像 +我们把PaddlePaddle的交叉编译环境打包成一个镜像，称为开发镜像，里面涵盖了交叉编译Android版PaddlePaddle库需要的所有编译工具。 + +```bash +$ git clone https://github.com/PaddlePaddle/Paddle.git +$ cd Paddle +$ docker build -t username/paddle-android:dev . -f Dockerfile.android +``` + +### 编译PaddlePaddle C-API库 +构建好开发镜像后，即可使用开发镜像来编译Android版PaddlePaddle C-API库。 +Android的Docker开发镜像向用户提供两个可配置的参数： + +| Argument | Optional Values | Default | +|-----------------|-------------------------|---------| +|`ANDROID_ABI` |`armeabi-v7a, arm64-v8a` | `armeabi-v7a` | +|`ANDROID_API` |`>= 21` | `21` | + +- 编译`armeabi-v7a`，`Android API 21`的PaddlePaddle库 +```bash +$ docker run -it --rm -v $PWD:/paddle -e "ANDROID_ABI=armeabi-v7a" -e "ANDROID_API=21" username/paddle-android:dev +``` + +- 编译`arm64-v8a`，`Android API 21`的PaddlePaddle库 +```bash +$ docker run -it --rm -v $PWD:/paddle -e "ANDROID_ABI=arm64-v8a" -e "ANDROID_API=21" username/paddle-android:dev +``` + +执行上述`docker run`命令时，容器默认执行[paddle/scripts/docker/build_android.sh](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/scripts/docker/build_android.sh)脚本。该脚本中记录了交叉编译Android版PaddlePaddle库常用的CMake配置，并且会根据`ANDROID_ABI`和`ANDROID_API`自动构建独立工具链、进行编译和安装。由于arm64架构要求Android API不小于21。因此当`ANDROID_ABI=arm64-v8a`，`ANDROID_API<21`时，Docker容器中将默认使用`Android API 21`的编译工具链。用户可以参考下文**配置交叉编译参数**章节，根据个人的需求修改定制Docker容器所执行的脚本。编译安装结束之后，PaddlePaddle的C-API库将被安装到`$PWD/install_android`目录，所依赖的第三方库同时也被安装到`$PWD/install_android/third_party`目录。 + +## 基于Linux交叉编译环境的编译方式 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 +### 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： @@ -13,18 +50,27 @@ 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： + +- 构建`armeabi-v7a`、 `Android API 21`的独立工具链： ```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 + --arch=arm --platform=android-21 --install-dir=your/path/to/arm_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。 +此命令将在`your/path/to/arm_standalone_toolchain`目录生成一套独立编译工具链，面向架构为32位ARM架构，支持的最小的Android API级别为21，支持编译器`arm-linux-androideabi-gcc (GCC) 4.9`和`clang 3.8`。 -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**。 +- 构建`arm64-v8a`、 `Android API 21`的独立工具链： +```bash +your/path/to/android-ndk-r14b-linux-x86_64/build/tools/make-standalone-toolchain.sh \ + --arch=arm64 --platform=android-21 --install-dir=your/path/to/arm64_standalone_toolchain +``` -## 配置交叉编译参数 +此命令将在`your/path/to/arm64_standalone_toolchain`目录生成一套独立编译工具链，面向架构为64位ARM64架构，支持的最小Android API级别为21，支持编译器`arm-linux-androideabi-gcc (GCC) 4.9`和`clang 3.8`。 + +注意：**PaddlePaddle要求使用的编译工具链所支持的Android 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)。 @@ -36,32 +82,57 @@ CMake系统对交叉编译提供了支持[cmake-toolchains](https://cmake.org/cm 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_TOOLCHAIN`，目标工具链。可设置`gcc/clang`，默认值为`clang`。 + - CMake 3.7以上，将会始终使用`clang`工具链；CMake 3.7以下，可设置`ANDROID_TOOLCHAIN=gcc`以使用`gcc`工具链。 + - Android官方提供的`clang`编译器要求系统支持`GLIBC 2.15`以上。 +- `ANDROID_ABI`，目标架构ABI。目前支持`armeabi-v7a`和`arm64-v8a`，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`。 +- `ANROID_ARM_MODE`，是否使用ARM模式。 + - `ANDROID_ABI=armeabi-v7a`时，可设置`ON/OFF`，默认值为`ON`； + - `ANDROID_ABI=arm64-v8a`时，不需要设置。 +- `ANDROID_ARM_NEON`，是否使用NEON指令。 + - `ANDROID_ABI=armeabi-v7a`时，可设置`ON/OFF`，默认值为`ON`； + - `ANDROID_ABI=arm64-v8a`时，不需要设置。 其他配置参数： +- `USE_EIGEN_FOR_BLAS`，是否使用Eigen库进行矩阵计算。可设置`ON/OFF`，默认值为`OFF`。 - `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： +常用的cmake配置如下： ```bash cmake -DCMAKE_SYSTEM_NAME=Android \ - -DANDROID_STANDALONE_TOOLCHAIN=your/path/to/my_standalone_toolchain \ + -DANDROID_STANDALONE_TOOLCHAIN=your/path/to/arm_standalone_toolchain \ -DANDROID_ABI=armeabi-v7a \ -DANDROID_ARM_NEON=ON \ -DANDROID_ARM_MODE=ON \ + -DUSE_EIGEN_FOR_BLAS=ON \ -DCMAKE_INSTALL_PREFIX=your/path/to/install \ -DWITH_C_API=ON \ -DWITH_SWIG_PY=OFF \ .. ``` +``` +cmake -DCMAKE_SYSTEM_NAME=Android \ + -DANDROID_STANDALONE_TOOLCHAIN=your/path/to/arm64_standalone_toolchain \ + -DANDROID_ABI=arm64-v8a \ + -DUSE_EIGEN_FOR_BLAS=OFF \ + -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 +**性能TIPS**，为了达到最快的计算速度，在CMake参数配置上，有以下建议： +- 设置`CMAKE_BUILD_TYPE`为`Release` +- 使用`clang`编译工具链 +- `armeabi-v7a`时，设置`USE_EIGEN_BLAS=ON`，使用Eigen进行矩阵计算；`arm64-v8a`时，设置`USE_EIGEN_FOR_BLAS=OFF`，使用OpenBLAS进行矩阵计算 + +### 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。 @@ -72,4 +143,4 @@ 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。 +执行完安装命令后，`your/path/to/install`目录中会包含`include`、`lib`和`third_party`目录，其中`include`中包含C-API的头文件，`lib`中包含若干个不同Android ABI的PaddlePaddle库，`third_party`中包含所依赖的所有第三方库。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/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