Multi-decoder setup for multi-task learning

[guillaumekln]

Similarly to the ParallelEncoder, a ParallelDecoder setup could allow multi-task learning. This should not be too hard to implement but we need to take care of some details:

• support separate values for the decoding parameters (beam_width, length_penalty, etc.),
• parts of SequenceToSequence assume a single output head (e.g. loss computation, reverse vocabulary lookup, exported outputs for model serving, etc. which should be moved in the decoder itself)

[BigFishMaster]

@guillaumekln Do you have a plan to implement ParallelDecoder?

[guillaumekln]

I don't have plan to work on this at the moment.

[sarubi]

@guillaumekln is there any possibility on OpenNMT-py to integrate multi encoder decoders setup for multi-task learning? If so I wish to contribute for it.

[guillaumekln]

This issue is for OpenNMT-tf. If you wish to contribute the feature for OpenNMT-py, can you update the issue you opened there?

[sarubi]

@guillaumekln updated on OpenNMT-py.