解决tf2 : AttributeError: 'Tensor' object has no attribute 'numpy' 笔记

问题描述：
【解决tf2 : AttributeError: 'Tensor' object has no attribute 'numpy'】环境：tf2.0
查看代码运行中tensor的值，

def create_mask(inputs, targets): input_leave_fun('create_mask',1) # 编码器只有padding_mask print("查看输入") print(type(inputs))print(inputs.numpy())# 此处报错

解决方法
注释@tf.function

#@tf.function注释此处 def train_step(inputs, targets): input_leave_fun('train_step',1) tar_inp = targets[:, :-1] tar_real = targets[:, 1:] print('tar_inp.shape {} tar_real.shape{}'.format(tar_inp.shape, tar_real.shape)) # 构造mask，调用函数处 encoder_padding_mask, look_ahead_mask, decoder_padding_mask = create_mask(inputs, tar_inp)with tf.GradientTape() as tape: pred, _ = transformer(inputs, tar_inp, True, encoder_padding_mask, look_ahead_mask, decoder_padding_mask) loss = loss_fun(tar_real, pred)print("执行退出") sys.exit(0)