[Bug]: 训练uie模型时报错ValueError: too many values to unpack (expected 2)

[yuzh0816]

软件环境

- paddlepaddle-gpu: 3.1.0
- paddlenlp: 3.0.0b4

重复问题

• I have searched the existing issues

错误描述

报错如下

Traceback (most recent call last):
  File "/root/PaddleNLP/slm/applications/information_extraction/document/evaluate.py", line 150, in <module>
    do_eval()
  File "/root/PaddleNLP/slm/applications/information_extraction/document/evaluate.py", line 113, in do_eval
    eval_metrics = trainer.evaluate()
                   ^^^^^^^^^^^^^^^^^^
  File "/root/miniconda3/envs/0708/lib/python3.11/site-packages/paddlenlp/trainer/trainer.py", line 3118, in evaluate
    output = self.evaluation_loop(
             ^^^^^^^^^^^^^^^^^^^^^
  File "/root/miniconda3/envs/0708/lib/python3.11/site-packages/paddlenlp/trainer/trainer.py", line 3340, in evaluation_loop
    metrics = self.compute_metrics(EvalPrediction(predictions=all_preds, label_ids=batch_labels))
              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/root/miniconda3/envs/0708/lib/python3.11/site-packages/paddlenlp/utils/ie_utils.py", line 134, in compute_metrics
    start_ids, end_ids = p.label_ids
    ^^^^^^^^^^^^^^^^^^
ValueError: too many values to unpack (expected 2)

稳定复现步骤 & 代码

训练UIE模型后，执行python evaluate.py --device "gpu" --model_path ./checkpoint/model_best --test_path ./data/dev.txt --output_dir ./checkpoint/model_best --label_names 'start_positions' 'end_positions' --max_seq_len 512 --per_device_eval_batch_size 16 --debug True进行训练后，产生报错。训练过程中没有报错。在evaluate时才出现报错。另外，在训练时加入do_eval也会产生报错。
参考文档：https://blog.csdn.net/z5z5z5z56/article/details/130346646

[yuzh0816]

如果使用PaddlePaddle 2.4.2 + PaddleNLP 2.5.2，则会报错：

Traceback (most recent call last):
  File "/root/PaddleNLP/slm/applications/information_extraction/document/information_extraction/document/evaluate.py", line 153, in <module>
    do_eval()
  File "/root/PaddleNLP/slm/applications/information_extraction/document/information_extraction/document/evaluate.py", line 84, in do_eval
    tokenizer = AutoTokenizer.from_pretrained(model_args.model_path)
  File "/root/miniconda3/envs/old/lib/python3.9/site-packages/paddlenlp/transformers/auto/tokenizer.py", line 333, in from_pretrained
    return tokenizer_class.from_pretrained(pretrained_model_name_or_path, *model_args, **kwargs)
  File "/root/miniconda3/envs/old/lib/python3.9/site-packages/paddlenlp/transformers/tokenizer_utils_base.py", line 1600, in from_pretrained
    value = AddedToken(**value)
TypeError: __init__() got an unexpected keyword argument 'special'

[Kongkong31415926]

软件环境

• paddlepaddle-gpu: 3.1.0
• paddlenlp: 3.0.0b4

重复问题

• I have searched the existing issues

错误描述

报错如下

Traceback (most recent call last):
File "/root/PaddleNLP/slm/applications/information_extraction/document/evaluate.py", line 150, in
do_eval()
File "/root/PaddleNLP/slm/applications/information_extraction/document/evaluate.py", line 113, in do_eval
eval_metrics = trainer.evaluate()
^^^^^^^^^^^^^^^^^^
File "/root/miniconda3/envs/0708/lib/python3.11/site-packages/paddlenlp/trainer/trainer.py", line 3118, in evaluate
output = self.evaluation_loop(
^^^^^^^^^^^^^^^^^^^^^
File "/root/miniconda3/envs/0708/lib/python3.11/site-packages/paddlenlp/trainer/trainer.py", line 3340, in evaluation_loop
metrics = self.compute_metrics(EvalPrediction(predictions=all_preds, label_ids=batch_labels))
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/root/miniconda3/envs/0708/lib/python3.11/site-packages/paddlenlp/utils/ie_utils.py", line 134, in compute_metrics
start_ids, end_ids = p.label_ids
^^^^^^^^^^^^^^^^^^
ValueError: too many values to unpack (expected 2)

稳定复现步骤 & 代码

训练UIE模型后，执行python evaluate.py --device "gpu" --model_path ./checkpoint/model_best --test_path ./data/dev.txt --output_dir ./checkpoint/model_best --label_names 'start_positions' 'end_positions' --max_seq_len 512 --per_device_eval_batch_size 16 --debug True进行训练后，产生报错。训练过程中没有报错。在evaluate时才出现报错。另外，在训练时加入do_eval也会产生报错。 参考文档：https://blog.csdn.net/z5z5z5z56/article/details/130346646

该报错是因为在134对一个ndarray对象进行解包报错

p对象是在3341行传入的实例化后的EvalPrediction对象。解包的label_ids属性就是EvalPrediction对象在实例化时传入的label_ids参数。在3339行对变量进行判断类型，若为list或tuple则取出第0个元素，这将会导致最终在解包时发生报错。我尝试注释掉该行，就成功解包。

仅供参考...

[lovely-fafa]

牛批，改了之后代码跑通了

[yuzh0816]

软件环境

• paddlepaddle-gpu: 3.1.0
• paddlenlp: 3.0.0b4

重复问题

• I have searched the existing issues

错误描述

报错如下
Traceback (most recent call last):
File "/root/PaddleNLP/slm/applications/information_extraction/document/evaluate.py", line 150, in
do_eval()
File "/root/PaddleNLP/slm/applications/information_extraction/document/evaluate.py", line 113, in do_eval
eval_metrics = trainer.evaluate()
^^^^^^^^^^^^^^^^^^
File "/root/miniconda3/envs/0708/lib/python3.11/site-packages/paddlenlp/trainer/trainer.py", line 3118, in evaluate
output = self.evaluation_loop(
^^^^^^^^^^^^^^^^^^^^^
File "/root/miniconda3/envs/0708/lib/python3.11/site-packages/paddlenlp/trainer/trainer.py", line 3340, in evaluation_loop
metrics = self.compute_metrics(EvalPrediction(predictions=all_preds, label_ids=batch_labels))
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/root/miniconda3/envs/0708/lib/python3.11/site-packages/paddlenlp/utils/ie_utils.py", line 134, in compute_metrics
start_ids, end_ids = p.label_ids
^^^^^^^^^^^^^^^^^^
ValueError: too many values to unpack (expected 2)

稳定复现步骤 & 代码

训练UIE模型后，执行python evaluate.py --device "gpu" --model_path ./checkpoint/model_best --test_path ./data/dev.txt --output_dir ./checkpoint/model_best --label_names 'start_positions' 'end_positions' --max_seq_len 512 --per_device_eval_batch_size 16 --debug True进行训练后，产生报错。训练过程中没有报错。在evaluate时才出现报错。另外，在训练时加入do_eval也会产生报错。 参考文档：https://blog.csdn.net/z5z5z5z56/article/details/130346646

该报错是因为在134对一个ndarray对象进行解包报错 p对象是在3341行传入的实例化后的EvalPrediction对象。解包的label_ids属性就是EvalPrediction对象在实例化时传入的label_ids参数。在3339行对变量进行判断类型，若为list或tuple则取出第0个元素，这将会导致最终在解包时发生报错。我尝试注释掉该行，就成功解包。 仅供参考...

@Kongkong31415926 万分感谢您提供的解决方案！
@gongel 问题确实出在这里，已经过验证，修改后能避免报错。本行判断是否有必要？若无必要，我这里提个pr把这行删掉。

[FT-Fetters]

nb，解决了