[Bug]: The file path "examples/text_to_knowledge/nptag/deploy/python/predict.py" may encounter an IndexError.

[smallbenxiong]

软件环境

- paddlepaddle:2.6.1
- paddlepaddle-gpu: 
- paddlenlp: 2.7.0

重复问题

• [X] I have searched the existing issues

错误描述

1. In the script:

   for i, d in enumerate(data):
   label = decode(pred_ids[i], id_vocabs)
   result = {
       "text": d,
       "label": label,
   }
   if label not in name_dict:
       scores_can = all_scores_can[i]
       pred_ids_can = all_preds_can[i]
       labels_can = search(scores_can, pred_ids_can, 0, [], 0)
       labels_can.sort(key=lambda d: -d[1])
       for labels in labels_can:
           cls_label_can = decode(labels[0], id_vocabs)
           if cls_label_can in name_dict:
               result["label"] = cls_label_can
               break
           else:
               labels_can = bk_tree.search_similar_word(label)
               result["label"] = labels_can[0][0]
   
   result["category"] = name_dict[result["label"]]
   results.append(result)
   return results

If labels_can is empty, accessing labels_can[0][0] may raise an IndexError.

Perhaps this is a way of modification:

labels_can = bk_tree.search_similar_word(label)
if len(labels_can) != 0:
    result["label"] = labels_can[0][0]

2.

result["category"] = name_dict[result["label"]]

If the issue in 1 is resolved, a KeyError might still occur in this section of the script.

稳定复现步骤 & 代码

Training with a small sample and adjusting the mapping JSON may lead to issues.

[wawltor]

Thank you for reporting the bug issue. You are welcome to create a pull request to address the bug based on your preferred method of modification.

[wawltor]

We made some changes in directory structure, the new path https://github.com/PaddlePaddle/PaddleNLP/tree/develop/legacy/examples/text_to_knowledge

[smallbenxiong]

Thank you for reporting the bug issue. You are welcome to create a pull request to address the bug based on your preferred method of modification.

Thank you for your trust. I will submit my solution code in the coming days.