Update train_multi_dcu.py

train_multi_dcu.py

@@ -104,7 +104,7 @@ def test_loop(dataloader, model):
         batch_data = {k: v.to(device) for k, v in batch_data.items()}
         with torch.no_grad():
             # 如果你使用了 DataParallel，你可以通过访问 model.module 来获取原始模型
-            generated_tokens = model.generate(
+            generated_tokens = model.module.generate(
                 batch_data["input_ids"],
                 attention_mask=batch_data["attention_mask"],
                 max_length=max_target_length,