Support for Multi Task Learning

[sarubi]

It would be great if OpenNMT-py can support multi-task learning such as having multiple encoder/decoders or else allow to share encoders with one or more different decoders or wise versa. As per my finding, in OpenNMT-tf this has been requested [1], [2] also we already have an open issue in OpenNMT-tf side. Do you have this feature included in the future road map of OpenNMT-py?

[1] https://forum.opennmt.net/t/opennmt-tf-multitask-learning-support/933 [2] https://forum.opennmt.net/t/multi-task-learning-after-the-encoded-layer/686

[sarubi]

I wish to contribute to this, is there a possibility to integrate multi encoder decoders setup for multi-task learning? since I'm new can anyone point out any docs or links which I can refer as a starting point?