recurrent_group cannot return more than one layer as its outputs. #2834
Description
I use machine translation in PaddleBook as the test example.
- Previously,
recurrent_groupin PaddlePaddle can return more than one layers inside the layer group as its outputs. - The returned layers are accessible to any other layers in the network, while layers inside
recurrent_group(and is not returned byrecurrent_gropu) are not accessible outside therecurrent_group. - I hope to use the above feature, but if I change this line to return more than one layers in
recurrent_group, the configuration parsing process crashes. It seems that there is some bug? - the error information is as follows:
Traceback (most recent call last): File "network_conf.py", line 114, in <module> print parse_network(seq_to_seq_net(30000, 30000, True, 3, 250)) File "network_conf.py", line 108, in seq_to_seq_net max_length=max_length) File "/home/caoying/paddle_codes/models/nmt_plot_attention_weights/paddle/v2/config_base.py", line 52, in wrapped out = f(*args, **xargs) File "/home/caoying/paddle_codes/models/nmt_plot_attention_weights/paddle/trainer_config_helpers/default_decorators.py", line 53, in __wrapper__ return func(*args, **kwargs) File "/home/caoying/paddle_codes/models/nmt_plot_attention_weights/paddle/trainer_config_helpers/layers.py", line 3949, in beam_search is_generating=True) File "/home/caoying/paddle_codes/models/nmt_plot_attention_weights/paddle/trainer_config_helpers/default_decorators.py", line 53, in __wrapper__ return func(*args, **kwargs) File "/home/caoying/paddle_codes/models/nmt_plot_attention_weights/paddle/trainer_config_helpers/layers.py", line 3657, in recurrent_group RecurrentLayerGroupEnd(name=name) File "/home/caoying/paddle_codes/models/nmt_plot_attention_weights/paddle/trainer/config_parser.py", line 396, in RecurrentLayerGroupEnd "memory declare wrong size:%d" % memory_link.size) File "/home/caoying/paddle_codes/models/nmt_plot_attention_weights/paddle/trainer/config_parser.py", line 156, in config_assert logger.fatal(msg) File "/home/caoying/paddle_codes/models/nmt_plot_attention_weights/paddle/trainer/config_parser.py", line 3531, in my_fatal raise Exception() Exception