examples/information_extraction/waybill_ie/run_bigru_crf.py 使用save_pretrained保存模型时报错

由于是本地使用，我修改了代码：使ErnieCrfForTokenClassification继承自ErniePretrainedModel

```
class ErnieCrfForTokenClassification(ErniePretrainedModel):
    def __init__(self, ernie, crf_lr=100):
        super(ErnieCrfForTokenClassification, self).__init__()
        self.num_classes = ernie.num_classes
        self.ernie = ernie  # allow ernie to be config
        self.crf = LinearChainCrf(
            self.num_classes, crf_lr=crf_lr, with_start_stop_tag=False)
        self.crf_loss = LinearChainCrfLoss(self.crf)
        self.viterbi_decoder = ViterbiDecoder(
            self.crf.transitions, with_start_stop_tag=False)
```

然后在保存模型时，使用model.save_pretrained('./ernie_crf_ckpt')保存模型，代码：
   
```
 step = 0
    best_f1 = 0
    for epoch in range(3000):
        for input_ids, token_type_ids, lengths, labels in train_loader:
            loss = model(
                input_ids, token_type_ids, lengths=lengths, labels=labels)
            avg_loss = paddle.mean(loss)
            avg_loss.backward()
            optimizer.step()
            optimizer.clear_grad()
            step += 1
            print("[TRAIN] Epoch:%d - Step:%d - Loss: %f" %
                  (epoch, step, avg_loss))
        f1_score = evaluate(model, metric, dev_loader)
        if f1_score >= best_f1:
            best_f1 = f1_score
            paddle.save(model.state_dict(), 'ernie_crf_ckpt/model_state.pdparams')
            # paddle.save(model, 'ernie_crf_ckpt/model_state.pdmodel')
            tokenizer.save_pretrained('./ernie_crf_ckpt')
            model.save_pretrained('./ernie_crf_ckpt')
```

出现报错：

```
Traceback (most recent call last):
  File "/home/user/Desktop/on_going_projects/sgaqfh/PaddleNLP/examples/information_extraction/waybill_ie/run_ernie_crf.py", line 135, in <module>
    model.save_pretrained('./ernie_crf_ckpt')
  File "/home/user/Desktop/on_going_projects/sgaqfh/PaddleNLP/paddlenlp/transformers/model_utils.py", line 386, in save_pretrained
    self.save_model_config(save_dir)
  File "/home/user/Desktop/on_going_projects/sgaqfh/PaddleNLP/paddlenlp/transformers/model_utils.py", line 374, in save_model_config
    f.write(json.dumps(model_config, ensure_ascii=False))
  File "/home/user/anaconda3/envs/paddle_2.1/lib/python3.8/json/__init__.py", line 234, in dumps
    return cls(
  File "/home/user/anaconda3/envs/paddle_2.1/lib/python3.8/json/encoder.py", line 199, in encode
    chunks = self.iterencode(o, _one_shot=True)
  File "/home/user/anaconda3/envs/paddle_2.1/lib/python3.8/json/encoder.py", line 257, in iterencode
    return _iterencode(o, 0)
  File "/home/user/anaconda3/envs/paddle_2.1/lib/python3.8/json/encoder.py", line 179, in default
    raise TypeError(f'Object of type {o.__class__.__name__} '
TypeError: Object of type ErnieModel is not JSON serializable
```

经过查看，是在保存model_config.json时报错，无法解决，求解决办法

PaddlePaddle/PaddleNLP

Content Risk Flag

Comments (7)

PaddlePaddle/PaddleNLP .gitee-modal { width: 500px !important; }

Content Risk Flag