基于seq2seq模型的简单对话系统的tf实现，具有embedding、attention、beam_search等功能，数据集是Cornell Movie Dialogs
143
stars
64
forks
source link
InvalidArgumentError (see above for traceback): Found Inf or NaN global norm. : Tensor had NaN values #21
Open
aguang1201 opened 5 years ago
每次训练到1000step的时候就报这个错误. 出错是model.py中,下面代码的最后一句: optimizer = tf.train.AdamOptimizer(self.learing_rate) trainable_params = tf.trainable_variables() gradients = tf.gradients(self.loss, trainable_params) _clipgradients, = tf.clip_by_global_norm(gradients, self.max_gradientnorm) 貌似是梯度消失吧.把学习率从0.0001改成0.001还是报错. 求指导