Created by: Kayven

对于输入【batch_size, seq_len, emb_len】 只能先fc成二维的【batch_size, hid_dim】，然后调用lstm？ emb=fluid.layers.fc(input=emb, size=hid_dim * 4) fluid.layers.dynamic_lstm(input=emb, size=hid_dim*4, is_reverse=False)

希望lstm输出【batch_size, seq_len, hid_dim】，fluid该怎么实现呢？ tensorflow里就是这样的，tf.nn.bidirectional_dynamic_rnn paddle里原来v2有个networks.bidirectional_lstm(..., return_seq=True)也可以实现，fluid不兼容