The Targets when they first get sent to the majority of models are of shape (Batch Size, Number Targets, Target Sequence Length), this is not compatible with the ELMO/BERT type encoder as they expect to have shape (Batch Size, Sequence Length) tensor as input.
The Targets when they first get sent to the majority of models are of shape (Batch Size, Number Targets, Target Sequence Length), this is not compatible with the ELMO/BERT type encoder as they expect to have shape (Batch Size, Sequence Length) tensor as input.