Update single_dcu_train.py

single_dcu_train.py

@@ -156,10 +156,10 @@ if __name__=='__main__':
         # 如果不存在，则创建文件夹
         os.makedirs(folder_path)
 
-    train_data = LCSTS('/umt5/data/lcsts_tsv/data1.tsv')
-    valid_data = LCSTS('/umt5/data/lcsts_tsv/data2.tsv')
+    train_data = LCSTS('./data/lcsts_tsv/data1.tsv')
+    valid_data = LCSTS('./data/lcsts_tsv/data2.tsv')
 
-    model_checkpoint = "/umt5/umt5_base"
+    model_checkpoint = "./umt5_base"
     tokenizer = AutoTokenizer.from_pretrained(model_checkpoint)
     model = AutoModelForSeq2SeqLM.from_pretrained(model_checkpoint)
 
@@ -194,7 +194,7 @@ if __name__=='__main__':
         if rouge_avg > best_avg_rouge:
             best_avg_rouge = rouge_avg
             print('saving new weights...\n')
-            weight_path = f'/utm5/saves/train_dtk_weights/epoch_{t+1}_valid_rouge_{rouge_avg:0.4f}_model_dtk_weights.bin'
+            weight_path = f'./saves/train_dtk_weights/epoch_{t+1}_valid_rouge_{rouge_avg:0.4f}_model_dtk_weights.bin'
             torch.save(model.state_dict(), weight_path)
             # 加载训练后的权重
             state_dict = torch.load(weight_path)