yongzhuo
commented
1 year ago 用m00_Bert，m14_transformer是初始的transformer的encode，没有适配预训练模型
Closed anglgn closed 1 year ago
yongzhuo
commented
1 year ago 用m00_Bert，m14_transformer是初始的transformer的encode，没有适配预训练模型
大佬，能否帮忙解决一些问题。使用的预训练模型是在github找的chinese_L-12_H-768_A-12，预训练模型加载成功，数据也正常处理了，但是出现了报错的情况如下： Traceback (most recent call last): File "E:/w_learning/KerasTC/keras_textclassification/m14_Transformer/train.py", line 98, in
train(rate=1)
File "E:/w_learning/KerasTC/keras_textclassification/m14_Transformer/train.py", line 93, in train
graph.fit(x_train, y_train, x_val, y_val)
File "E:\w_learning\KerasTC\keras_textclassification\base\graph.py", line 151, in fit
callbacks=self.callback())
File "G:\anaconda\envs\KerasTC\lib\site-packages\keras\engine\training.py", line 952, in fit
batch_size=batch_size)
File "G:\anaconda\envs\KerasTC\lib\site-packages\keras\engine\training.py", line 751, in _standardize_user_data
exception_prefix='input')
File "G:\anaconda\envs\KerasTC\lib\site-packages\keras\engine\training_utils.py", line 102, in standardize_input_data
str(len(data)) + ' arrays: ' + str(data)[:200] + '...')
ValueError: Error when checking model input: the list of Numpy arrays that you are passing to your model is not the size the model expected. Expected to see 1 array(s), but instead got the following list of 2 arrays: [array([[ 101, 4510, 928, ..., 0, 0, 0],
[ 101, 4956, 1762, ..., 0, 0, 0],
[ 101, 7028, 2412, ..., 0, 0, 0],
...,
[ 101, 3300, 1068, ..., 0, ...
初步排查好像是text_preprocess.py里的第281行
if embeddingtype in ['bert', 'albert']:
x, y_ = np.array(x), np.array(label_zo)
x1 = np.array([x[0] for x in x])
x2 = np.array([x[1] for x in x])
x_all = [x_1, x_2]
return xall, y
出现了问题。问题描述为输入的数组为2个，而模型只能输入一个。于是测试了一下只输入 x_1或x_2模型可以模型起来，但是数据不完全，准确率很低。大佬，想问下这里该怎么改呀。