confirmed for 8vcpu

new_test_saved_finetuned_model.py

@@ -46,16 +46,16 @@ class BERTFineTuneTrainer:
         """
 
         # Setup cuda device for BERT training, argument -c, --cuda should be true
-        cuda_condition = torch.cuda.is_available() and with_cuda
-        self.device = torch.device("cuda:0" if cuda_condition else "cpu")
-        #self.device = torch.device("cpu") #torch.device("cuda:0" if cuda_condition else "cpu")
+        # cuda_condition = torch.cuda.is_available() and with_cuda
+        # self.device = torch.device("cuda:0" if cuda_condition else "cpu")
+        self.device = torch.device("cpu") #torch.device("cuda:0" if cuda_condition else "cpu")
         # print(cuda_condition, " Device used = ", self.device)
         print(" Device used = ", self.device)
         
         # available_gpus = list(range(torch.cuda.device_count()))
 
         # This BERT model will be saved every epoch
-        self.model = bertFinetunedClassifierwithFeats.to(self.device)
+        self.model = bertFinetunedClassifierwithFeats.to("cpu")
         print(self.model.parameters())
         for param in self.model.parameters():
             param.requires_grad = False
@@ -159,11 +159,10 @@ class BERTFineTuneTrainer:
                     logits = self.model.forward(data["input"], data["segment_label"], data["feat"])
                 else:
                     with torch.no_grad():
-                        logits = self.model.forward(data["input"], data["segment_label"], data["feat"])
+                        logits = self.model.forward(data["input"].cpu(), data["segment_label"].cpu(), data["feat"].cpu())
 
                 logits = logits.cpu()
-                labels = data["label"].to(logits.device)
-                loss = self.criterion(logits, labels)
+                loss = self.criterion(logits, data["label"])
                 # if torch.cuda.device_count() > 1:
                 #     loss = loss.mean()
 
@@ -183,8 +182,7 @@ class BERTFineTuneTrainer:
                 tlabels.extend(data['label'].cpu().numpy())
                 positive_class_probs = [prob[1] for prob in probabs]
                 # Compare predicted labels to true labels and calculate accuracy
-                correct = (data['label'].to(predicted_labels.device) == predicted_labels).sum().item()
-
+                correct = (data['label'] == predicted_labels).sum().item()
                 
                 avg_loss += loss.item()
                 total_correct += correct
@@ -560,7 +558,7 @@ def train():
     parser.add_argument("-w", "--num_workers", type=int, default=7, help="dataloader worker size")
 
     # Later run with cuda
-    parser.add_argument("--with_cuda", type=bool, default=True, help="training with CUDA: true, or false")
+    parser.add_argument("--with_cuda", type=bool, default=False, help="training with CUDA: true, or false")
     parser.add_argument("--log_freq", type=int, default=10, help="printing loss every n iter: setting n")
     # parser.add_argument("--corpus_lines", type=int, default=None, help="total number of lines in corpus")
     parser.add_argument("--cuda_devices", type=int, nargs='+', default=None, help="CUDA device ids")