merge

README.md

@@ -1,8 +1,10 @@
 
 # AmuNMT
 [![Join the chat at https://gitter.im/amunmt/amunmt](https://badges.gitter.im/amunmt/amunmt.svg)](https://gitter.im/amunmt/amunmt?utm_source=badge&utm_medium=badge&utm_campaign=pr-badge&utm_content=badge)
-[![CUDA build Status](http://37.247.57.181:8000/job/amunmt_compilation_cuda/badge/icon)](http://37.247.57.181:8000/job/amunmt_compilation_cuda/)
+
-[![CPU Build Status](http://37.247.57.181:8000/job/amunmt_compilation_cpu/badge/icon)](http://37.247.57.181:8000/job/amunmt_compilation_cpu/)
+[![CUDABuild Status](http://vali.inf.ed.ac.uk/jenkins/buildStatus/icon?job=amunmt_compilation_cuda)](http://vali.inf.ed.ac.uk/jenkins/job/amunmt_compilation_cuda/)
+[![CPU Build Status](http://vali.inf.ed.ac.uk/jenkins/buildStatus/icon?job=amunmt_compilation_cpu)](http://vali.inf.ed.ac.uk/jenkins/job/amunmt_compilation_cpu/)
+
 
 A C++ inference engine for Neural Machine Translation (NMT) models trained with Theano-based scripts from
 Nematus (https://github.com/rsennrich/nematus) or DL4MT (https://github.com/nyu-dl/dl4mt-tutorial)
@@ -92,14 +94,35 @@ AmuNMT has integrated support for [BPE encoding](https://github.com/rsennrich/su
     bpe: bpe.codes
     debpe: true
 
+## Python Bindings
+
+Python bindings allow to run AmuNMT decoder in python scripts. The compilation of the bindings requires `python-dev` package. To compile the bindings run:
+```
+make python
+```
+
+The Python bindings consist of 2 function: `init` and `translate`:
+
+```python
+import libamunmt
+
+libamunmt.init('-c config.yml')
+print libamunmt.translate(['this is a little test .'])
+```
+
+The `init` function init the decoder and the syntax is the same as in command line. The `translate`
+function takes a list of sentences to translate. For real-world example, see the `scripts/amunmt_erver.py`
+script, which uses python bindings to run REST server.
+
+
 ## Using GPU/CPU threads
-AmuNMT can use GPUs, CPUs, or both, to distribute translation of different sentences. 
+AmuNMT can use GPUs, CPUs, or both, to distribute translation of different sentences. **However, it is unlikely that CPUs used together with GPUs yield any performance improvement. It is probably better to only use the GPU if one or more are available.**
 
     cpu-threads: 8
     gpu-threads: 2
     devices: [0, 1]
 
-The setting above uses 8 CPU threads and 4 GPU threads (2 GPUs x 2 threads). The `gpu-threads` and `devices` options are only available when AmuNMT has been compiled with CUDA support. Multiple GPU threads can be used to increase GPU saturation, but will likely not result in a large performance boost. By default, `gpu-threads` is set to `1` and `cpu-threads` to `0`  if CUDA is available. Otherwise `cpu-threads` is set to `1`. To disable the GPU set `gpu-threads` to `0`. Setting both `gpu-threads` and `cpu-threads` to `0` will result in an exception.
+The setting above uses 8 CPU threads and 4 GPU threads (2 GPUs x 2 threads). The `gpu-threads` and `devices` options are only available when AmuNMT has been compiled with CUDA support. Multiple GPU threads can be used to increase GPU saturation, but will likely not result in a large performance boost. By default, `gpu-threads` is set to `1` and `cpu-threads` to `0`  if CUDA is available. Otherwise `cpu-threads` is set to `1`. To disable the GPU set `gpu-threads` to `0`. Setting both `gpu-threads` and `cpu-threads` to `0` will result in an exception. 
 
 ## Example usage
 

src/CMakeLists.txt

@@ -64,7 +64,7 @@ cuda_add_executable(
 )
 
 if(PYTHONLIBS_FOUND)
-cuda_add_library(amunmt SHARED
+cuda_add_library(python SHARED
   python/amunmt.cpp
   # gpu/decoder/ape_penalty.cu
   gpu/decoder/encoder_decoder.cu
@@ -80,7 +80,9 @@ cuda_add_library(amunmt SHARED
   $<TARGET_OBJECTS:cpumode>
   $<TARGET_OBJECTS:libyaml-cpp>
 )
-set_target_properties("amunmt" PROPERTIES EXCLUDE_FROM_ALL 1)
+set_target_properties("python" PROPERTIES EXCLUDE_FROM_ALL 1)
+set_target_properties("python" PROPERTIES OUTPUT_NAME "amunmt")
+
 endif(PYTHONLIBS_FOUND)
 
 cuda_add_library(mosesplugin STATIC
@@ -116,7 +118,7 @@ add_executable(
 )
 
 if(PYTHONLIBS_FOUND)
-add_library(amunmt SHARED
+add_library(python SHARED
   python/amunmt.cpp
   common/loader_factory.cpp
   $<TARGET_OBJECTS:libcnpy>
@@ -124,14 +126,15 @@ add_library(amunmt SHARED
   $<TARGET_OBJECTS:libcommon>
   $<TARGET_OBJECTS:libyaml-cpp>
 )
-set_target_properties("amunmt" PROPERTIES EXCLUDE_FROM_ALL 1)
+set_target_properties("python" PROPERTIES EXCLUDE_FROM_ALL 1)
+set_target_properties("python" PROPERTIES OUTPUT_NAME "amunmt")
 endif(PYTHONLIBS_FOUND)
 endif(CUDA_FOUND)
 
 SET(EXES "amun")
 
 if(PYTHONLIBS_FOUND)
-SET(EXES ${EXES} "amunmt")
+SET(EXES ${EXES} "python")
 endif(PYTHONLIBS_FOUND)
 
 foreach(exec ${EXES})

src/common/god.cpp

@@ -22,20 +22,22 @@ using namespace std;
 namespace amunmt {
 
 God::God()
-:threadIncr_(0)
+ : threadIncr_(0)
 {
-
 }
 
-God::~God() {}
+God::~God()
+{
+}
 
 God& God::Init(const std::string& options) {
   std::vector<std::string> args = boost::program_options::split_unix(options);
   int argc = args.size() + 1;
   char* argv[argc];
   argv[0] = const_cast<char*>("bogus");
-  for(int i = 1; i < argc; i++)
+  for (int i = 1; i < argc; ++i) {
     argv[i] = const_cast<char*>(args[i-1].c_str());
+  }
   return Init(argc, argv);
 }
 
@@ -49,35 +51,35 @@ God& God::Init(int argc, char** argv) {
   config_.AddOptions(argc, argv);
   config_.LogOptions();
 
-  if(Get("source-vocab").IsSequence()) {
+  if (Get("source-vocab").IsSequence()) {
-	for(auto sourceVocabPath : Get<std::vector<std::string>>("source-vocab"))
+    for (auto sourceVocabPath : Get<std::vector<std::string>>("source-vocab")) {
-	  sourceVocabs_.emplace_back(new Vocab(sourceVocabPath));
+      sourceVocabs_.emplace_back(new Vocab(sourceVocabPath));
-  }
+    }
-  else {
+  } else {
-	sourceVocabs_.emplace_back(new Vocab(Get<std::string>("source-vocab")));
+    sourceVocabs_.emplace_back(new Vocab(Get<std::string>("source-vocab")));
   }
   targetVocab_.reset(new Vocab(Get<std::string>("target-vocab")));
 
   weights_ = Get<std::map<std::string, float>>("weights");
 
   if(Get<bool>("show-weights")) {
-	LOG(info) << "Outputting weights and exiting";
+    LOG(info) << "Outputting weights and exiting";
-	for(auto && pair : weights_) {
+    for(auto && pair : weights_) {
-	  std::cout << pair.first << "= " << pair.second << std::endl;
+      std::cout << pair.first << "= " << pair.second << std::endl;
-	}
+    }
-	exit(0);
+    exit(0);
   }
 
   LoadScorers();
   LoadFiltering();
 
   if (Has("input-file")) {
-	LOG(info) << "Reading from " << Get<std::string>("input-file");
+    LOG(info) << "Reading from " << Get<std::string>("input-file");
-	inputStream_.reset(new InputFileStream(Get<std::string>("input-file")));
+    inputStream_.reset(new InputFileStream(Get<std::string>("input-file")));
   }
   else {
-	LOG(info) << "Reading from stdin";
+    LOG(info) << "Reading from stdin";
-	inputStream_.reset(new InputFileStream(std::cin));
+    inputStream_.reset(new InputFileStream(std::cin));
   }
 
   LoadPrePostProcessing();
@@ -184,11 +186,9 @@ std::vector<ScorerPtr> God::GetScorers(const DeviceInfo &deviceInfo) const {
   std::vector<ScorerPtr> scorers;
 
   if (deviceInfo.deviceType == CPUDevice) {
-    //cerr << "CPU GetScorers" << endl;
     for (auto&& loader : cpuLoaders_ | boost::adaptors::map_values)
       scorers.emplace_back(loader->NewScorer(*this, deviceInfo));
   } else {
-    //cerr << "GPU GetScorers" << endl;
     for (auto&& loader : gpuLoaders_ | boost::adaptors::map_values)
       scorers.emplace_back(loader->NewScorer(*this, deviceInfo));
   }
@@ -233,14 +233,12 @@ std::vector<std::string> God::Postprocess(const std::vector<std::string>& input)
   }
   return processed;
 }
-// clean up cuda vectors before cuda context goes out of scope
+
 void God::CleanUp() {
   for (Loaders::value_type& loader : cpuLoaders_) {
-    //cerr << "cpu loader=" << loader.first << endl;
      loader.second.reset(nullptr);
   }
   for (Loaders::value_type& loader : gpuLoaders_) {
-    //cerr << "gpu loader=" << loader.first << endl;
      loader.second.reset(nullptr);
   }
 }
@@ -274,7 +272,6 @@ DeviceInfo God::GetNextDevice() const
 
   ++threadIncr_;
 
-  //cerr << "GetNextDevice=" << ret << endl;
   return ret;
 }