Python,快速专家,使用预先训练的模型进行预测

问题描述

我正在使用快速的Bert软件包来训练Bert模型。 快速的bert保存模型输出以下文件

-Resources
-- config.json
-- pytorch_model.bin
-- specail_tokens_map.json
-- tokenizer_config.json
-- vocab.txt
-- events.out.tfevents.1601151257 (1).ffce4853f2c9

我正尝试通过快速bert加载模型并使用它进行预测:

from fast_bert.prediction import BertClassificationPredictor
def predictor(texts,MODEL_PATH=None):
        """
        :param MODEL_PATH: path to trained model
        :type texts: list
        :param: texts: texts to run prediction on
        """
        MODEL_PATH = '/content/Resources/' 
        LABEL_DATA = '/content/data/' # same path as when trained so it's valid

        if MODEL_PATH is None:
            raise LookupError("This Path is either wrong or No Trained Model exists")
        predictor = BertClassificationPredictor(
            model_path=MODEL_PATH,label_path=LABEL_DATA,# location for labels.csv file
            multi_label=False,model_type='xlnet',do_lower_case=False)

        # Batch predictions
        multiple_predictions = predictor.predict_batch(texts)
        pprint(("Predicting texts accuracy",) multiple_predictions)
        return multiple_predictions

因此,出现以下错误

OSError: Unable to load weights from pytorch checkpoint file. If you tried to load a PyTorch model from a TF 2.0 checkpoint,please set from_tf=True. 

解决方法

暂无找到可以解决该程序问题的有效方法,小编努力寻找整理中!

如果你已经找到好的解决方法,欢迎将解决方案带上本链接一起发送给小编。

小编邮箱:dio#foxmail.com (将#修改为@)