Logs

app.py

@@ -151,8 +151,8 @@ def evaluate_model(model, dataloader, device):
             # Process each sequence in the batch
             for i in range(current_batch_size):
                 valid_mask = (labels[i] != -100) & (attention_mask[i] != 0)
-                valid_preds = preds[i][valid_mask].flatten()
-                valid_labels = labels[i][valid_mask].flatten()
+                valid_preds = preds[i][valid_mask[i]].flatten()
+                valid_labels = labels[i][valid_mask[i]].flatten()
                 print(valid_mask.dtype)
                 print(labels[i].shape)
                 print(labels[i])
@@ -160,11 +160,7 @@ def evaluate_model(model, dataloader, device):
                 print(valid_mask.shape)
                 print(valid_labels)
                 print(valid_mask)
-                all_preds.extend(valid_preds.tolist())
-                all_labels.extend(valid_labels.tolist())
-
                 assert not torch.any(valid_labels == -100), f"Found -100 in valid_labels for batch {i}"
-
                 if sample_count < num_samples:
                     print(f"Sample {sample_count + 1}:")
                     print(f"Tokens: {tokenizer.convert_ids_to_tokens(input_ids[i])}")
@@ -172,6 +168,8 @@ def evaluate_model(model, dataloader, device):
                     print(f"Predicted Labels: {[id2label[pred] for pred in valid_preds]}")
                     print("-" * 50)
                     sample_count += 1
+                all_preds.extend(valid_preds.tolist())
+                all_labels.extend(valid_labels.tolist())
 
     # Calculate evaluation metrics
     print("evaluate_model sizes")