原因是使用Adam优化函数时，Adam函数会创建一个Adam变量，目的是保存你使用tensorflow创建的graph中的每个可训练参数的动量，
words/_word_embeddings:0
bi-lstm/bidirectional_rnn/fw/lstm_cell/kernel:0
bi-lstm/bidirectional_rnn/fw/lstm_cell/bias:0
bi-lstm/bidirectional_rnn/bw/lstm_cell/kernel:0
bi-lstm/bidirectional_rnn/bw/lstm_cell/bias:0
proj/W:0
proj/b:0
bi-lstm_secondLayer/bidirectionalt_rnn/fw/lstm_cell/kernel:0
bi-lstm_secondLayer/bidirectional_rnn/fw/lstm_cell/bias:0
bi-lstm_secondLayer/bidirectional_rnn/bw/lstm_cell/kernel:0
bi-lstm_secondLayer/bidirectional_rnn/bw/lstm_cell/bias:0
proj_secondLayer/W:0
proj_secondLayer/b_secondLayer:0
train_step_secondLayer/beta1_power:0
train_step_secondLayer/beta2_power:0
train_step_secondLayer/words/_word_embeddings/Adam:0
train_step_secondLayer/words/_word_embeddings/Adam_1:0
train_step_secondLayer/bi-lstm/bidirectional_rnn/fw/lstm_cell/kernel/Adam:0
train_step_secondLayer/bi-lstm/bidirectional_rnn/fw/lstm_cell/kernel/Adam_1:0
train_step_secondLayer/bi-lstm/bidirectional_rnn/fw/lstm_cell/bias/Adam:0
train_step_secondLayer/bi-lstm/bidirectional_rnn/fw/lstm_cell/bias/Adam_1:0
train_step_secondLayer/bi-lstm/bidirectional_rnn/bw/lstm_cell/kernel/Adam:0
train_step_secondLayer/bi-lstm/bidirectional_rnn/bw/lstm_cell/kernel/Adam_1:0
train_step_secondLayer/bi-lstm/bidirectional_rnn/bw/lstm_cell/bias/Adam:0
train_step_secondLayer/bi-lstm/bidirectional_rnn/bw/lstm_cell/bias/Adam_1:0
train_step_secondLayer/proj/W/Adam:0
train_step_secondLayer/proj/W/Adam_1:0
train_step_secondLayer/proj/b/Adam:0
train_step_secondLayer/proj/b/Adam_1:0
train_step_secondLayer/bi-lstm_secondLayer/bidirectional_rnn/fw/lstm_cell/kernel/Adam:0
train_step_secondLayer/bi-lstm_secondLayer/bidirectional_rnn/fw/lstm_cell/kernel/Adam_1:0
train_step_secondLayer/bi-lstm_secondLayer/bidirectional_rnn/fw/lstm_cell/bias/Adam:0
train_step_secondLayer/bi-lstm_secondLayer/bidirectional_rnn/fw/lstm_cell/bias/Adam_1:0
train_step_secondLayer/bi-lstm_secondLayer/bidirectional_rnn/bw/lstm_cell/kernel/Adam:0
train_step_secondLayer/bi-lstm_secondLayer/bidirectional_rnn/bw/lstm_cell/kernel/Adam_1:0
train_step_secondLayer/bi-lstm_secondLayer/bidirectional_rnn/bw/lstm_cell/bias/Adam:0
train_step_secondLayer/bi-lstm_secondLayer/bidirectional_rnn/bw/lstm_cell/bias/Adam_1:0
train_step_secondLayer/proj_secondLayer/W/Adam:0
train_step_secondLayer/proj_secondLayer/W/Adam_1:0
train_step_secondLayer/proj_secondLayer/b_secondLayer/Adam:0
train_step_secondLayer/proj_secondLayer/b_secondLayer/Adam_1:0