updates

source/train/TTrain.cpp

@@ -80,18 +80,20 @@ void TestTrain()
     config.Add("lrate", 0.001F);
     config.Add("lrate", 0.001F);
     config.Add("nstep", 100000);
     config.Add("nstep", 100000);
     config.Add("nepoch", 5);
     config.Add("nepoch", 5);
-    config.Add("jobdev0", -1);
-    //config.Add("jobdev1", -1);
-    //config.Add("jobdev2", -1);
+    config.Add("jobdev0", 0);
+    config.Add("jobdev1", -1);
+    config.Add("jobdev2", -1);
     //config.Add("jobdev3", -1);
     //config.Add("jobdev3", -1);
     //config.Add("jobdev4", -1);
     //config.Add("jobdev4", -1);
 
 
+    int serverDevID = config.GetInt("jobdev0", -1);
+
     TTDataLoader loader;
     TTDataLoader loader;
     loader.SetFileName("ttrain.txt");
     loader.SetFileName("ttrain.txt");
     loader.SetBatchSize(config.GetInt("batchsize", TT_BATCH_SIZE));
     loader.SetBatchSize(config.GetInt("batchsize", TT_BATCH_SIZE));
 
 
     TTModel model;
     TTModel model;
-    model.Init(config, -1);
+    model.Init(config, serverDevID);
 
 
     tmpTT = model.params[0].param;
     tmpTT = model.params[0].param;
 
 
@@ -218,6 +220,10 @@ bool TTDataLoader::GetBatchSimple(XList * inputs, XList * golds)
 /* constructor */
 /* constructor */
 TTModel::TTModel()
 TTModel::TTModel()
 {
 {
+    devID = -1;
+    vSize = 0;
+    eSize = 0;
+    hSize = 0;
 }
 }
 
 
 /* de-constructor */
 /* de-constructor */
@@ -236,11 +242,13 @@ initialize the model
 >> myConfig - configuration
 >> myConfig - configuration
 >> devID - device id
 >> devID - device id
 */
 */
-void TTModel::Init(XConfig &myConfig, int devID)
+void TTModel::Init(XConfig &myConfig, int myDevID)
 {
 {
     Clear();
     Clear();
     SetConfig(myConfig);
     SetConfig(myConfig);
 
 
+    devID = myDevID;
+
     vSize = MAX_INT_IN_TTRAIN + 1;
     vSize = MAX_INT_IN_TTRAIN + 1;
     eSize = config.GetInt("esize", TT_EMBEDDING_SIZE);
     eSize = config.GetInt("esize", TT_EMBEDDING_SIZE);
     hSize = config.GetInt("hsize", TT_HIDDEN_SIZE);
     hSize = config.GetInt("hsize", TT_HIDDEN_SIZE);
@@ -331,6 +339,11 @@ bool TTModel::RunSimple(XList * inputs, XList * outputs, XList * golds, XList* l
     XTensor * loss = (XTensor*)losses->GetItem(0);
     XTensor * loss = (XTensor*)losses->GetItem(0);
     XTensor goldOneHot;
     XTensor goldOneHot;
 
 
+    /* place all input data on the correct device */
+    input->FlushToDevice(devID);
+    output->FlushToDevice(devID);
+    gold->FlushToDevice(devID);
+
     XNet net;
     XNet net;
 
 
     /* create the neural network and run it */
     /* create the neural network and run it */

source/train/TTrain.h

@@ -49,8 +49,8 @@ namespace nts { // namespace nts(NiuTrans.Tensor)
 #define MAX_SAMPLE_LINE_LENGTH 128
 #define MAX_SAMPLE_LINE_LENGTH 128
 #define MAX_SAMPLE_SIZE 4
 #define MAX_SAMPLE_SIZE 4
 #define TT_BATCH_SIZE 256
 #define TT_BATCH_SIZE 256
-#define TT_EMBEDDING_SIZE 64
-#define TT_HIDDEN_SIZE 256
+#define TT_EMBEDDING_SIZE 128
+#define TT_HIDDEN_SIZE 512
 
 
 extern XTensor * tmpTT;
 extern XTensor * tmpTT;
 
 
@@ -136,7 +136,7 @@ public:
     void SetConfig(XConfig &myConfig);
     void SetConfig(XConfig &myConfig);
 
 
     /* initialize the parameters */
     /* initialize the parameters */
-    void Init(XConfig &myConfig, int devID);
+    void Init(XConfig &myConfig, int myDevID);
 
 
     /* create the model */
     /* create the model */
     void Forward(int devID, XTensor * input, XTensor * output);
     void Forward(int devID, XTensor * input, XTensor * output);

source/train/XLeader.cpp

@@ -265,7 +265,7 @@ void XLeader::AddJobWorker(XModel * model, int n, int * ids)
     }
     }
 
 
     /* we clone the input model */
     /* we clone the input model */
-    for (int i = 0; i < n - 1; i++) {
+    for (int i = 1; i < n; i++) {
         XWorkerJob * worker = new XWorkerJob();
         XWorkerJob * worker = new XWorkerJob();
         worker->SetModel(model->Clone(ids[i]));
         worker->SetModel(model->Clone(ids[i]));
         jworkers.Add(worker);
         jworkers.Add(worker);

source/train/XOptimizer.cpp

@@ -63,9 +63,9 @@ void XOptimizer::Clear()
 void XOptimizer::ShowSettings()
 void XOptimizer::ShowSettings()
 {
 {
     XPRINT(1, stderr, "[INFO] Optimizer Setup:\n");
     XPRINT(1, stderr, "[INFO] Optimizer Setup:\n");
-    XPRINT1(1, stderr, "       nstep = %d\n", nstep);
-    XPRINT1(1, stderr, "       nepoch = %d\n", nepoch);
-    XPRINT1(1, stderr, "       lrate = %.3f\n", lrate);
+    XPRINT2(1, stderr, "%25s = %d\n", "nstep", nstep);
+    XPRINT2(1, stderr, "%25s = %d\n", "nepoch", nepoch);
+    XPRINT2(1, stderr, "%25s = %.3f\n", "lrate", lrate);
 }
 }
 
 
 /* 
 /* 

source/train/XTrainer.cpp

@@ -102,6 +102,7 @@ void XTrainer::Run(XConfig * config, DataDistributeBase * dataDistributor,
     GetDevIDs(config, ids, jobNum, MAX_DEVICE_NUM_TRAINING);
     GetDevIDs(config, ids, jobNum, MAX_DEVICE_NUM_TRAINING);
 
 
     optimizer->ShowSettings();
     optimizer->ShowSettings();
+    this->ShowSettings(config);
 
 
     /* create the server and workers */
     /* create the server and workers */
     XLeader leader;
     XLeader leader;
@@ -148,4 +149,27 @@ void XTrainer::Run(XConfig * config, DataDistributeBase * dataDistributor,
     delete[] ids;
     delete[] ids;
 }
 }
 
 
+/* show settings of training */
+void XTrainer::ShowSettings(XConfig* config)
+{
+    int workerNum = 0;
+    int* ids = new int[MAX_DEVICE_NUM_TRAINING];
+
+    GetDevIDs(config, ids, workerNum, MAX_DEVICE_NUM_TRAINING);
+
+    XPRINT(1, stderr, "[INFO] Training Setup:\n");
+    XPRINT2(1, stderr, "%25s = %d\n", "nworker", workerNum);
+
+    if (workerNum > 0) {
+        XPRINT2(1, stderr, "%25s = device[%d]\n", "worker0(server)", ids[0]);
+        for (int i = 1; i < workerNum; i++) {
+            char name[32];
+            sprintf(name, "worker%d", i);
+            XPRINT2(1, stderr, "%25s = device[%d]\n", name, ids[i]);
+        }
+    }
+
+    delete[] ids;
+}
+
 } /* end of the nts (NiuTrans.Tensor) namespace */
 } /* end of the nts (NiuTrans.Tensor) namespace */

source/train/XTrainer.h

@@ -77,6 +77,9 @@ public:
     virtual
     virtual
     void Run(XConfig * config, DataDistributeBase * dataDistributor, 
     void Run(XConfig * config, DataDistributeBase * dataDistributor, 
              XModel * model, XOptimizer * optimizer);
              XModel * model, XOptimizer * optimizer);
+
+    /* show settings of training */
+    void ShowSettings(XConfig * config);
 };
 };
 }
 }
 #endif // __XTRAINER_H__
 #endif // __XTRAINER_H__
\ No newline at end of file