Logs

app.py

@@ -123,13 +123,9 @@ def evaluate_model(model, dataloader, device):
 
     # Disable gradient calculations
     with torch.no_grad():
-        for batch in dataloader:
-            print("Sample sequence labels:", batch['labels'][0].tolist()[:20])
-            print("Corresponding predictions:", torch.argmax(model(batch['input_ids'].to(device), 
-                attention_mask=batch['attention_mask'].to(device)).logits, dim=-1)[0].tolist()[:20])
-            break
         for batch in dataloader:
             input_ids = batch['input_ids'].to(device)
+            current_batch_size = input_ids.size(0)
             attention_mask = batch['attention_mask'].to(device)
             labels = batch['labels'].to(device).cpu().numpy()
 
@@ -140,11 +136,15 @@ def evaluate_model(model, dataloader, device):
             # Get predictions
             preds = torch.argmax(logits, dim=-1).cpu().numpy()
 
-            print("Shape of preds:", preds.shape)
-            print("Shape of labels:", labels.shape)
+             # Use attention mask to get valid tokens
+            mask = batch['attention_mask'].cpu().numpy().astype(bool)
             
-            all_preds.extend(preds)
-            all_labels.extend(labels)
+            # Process each sequence in the batch
+            for i in range(current_batch_size):
+                valid_preds = preds[i][mask[i]].flatten()
+                valid_labels = labels[i][mask[i]].flatten()
+                all_preds.extend(valid_preds.tolist())
+                all_labels.extend(valid_labels.tolist())
 
     # Calculate evaluation metrics
     print("evaluate_model sizes")