Logs

app.py

@@ -137,7 +137,7 @@ def evaluate_model(model, dataloader, device):
         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)
+            attention_mask = batch['attention_mask'].cpu().numpy().astype(bool)
             labels = batch['labels'].to(device).cpu().numpy()
 
             # Forward pass to get logits
@@ -147,14 +147,12 @@ def evaluate_model(model, dataloader, device):
 
             # Get predictions
             preds = torch.argmax(logits, dim=-1).cpu().numpy()
-
-             # Use attention mask to get valid tokens
-            mask = batch['attention_mask'].cpu().numpy().astype(bool)
             
             # 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()
+                valid_mask = (labels[i] != -100) & attention_mask[i]
+                valid_preds = preds[i][valid_mask[i]].flatten()
+                valid_labels = labels[i][valid_mask[i]].flatten()
                 all_preds.extend(valid_preds.tolist())
                 all_labels.extend(valid_labels.tolist())