Merge branch 'michael_issue#110_parse_hyperparameters_to_dataset_instance' into develop

@langguth1 Now I remember why we use self.model_hparams_dict_load.update({"sequence_length": self.train_dataset.sequence_length}), because we remove the hyperperatmers "sequence_length" from dataset, we use the default sequence_length of the tensorRecords. When training model , we need to pass this information to the model part. We have to add it back. I think we revised this together :) should I comment on it?

Yes, I also remember now. I've also readded it in the current working branch #116 (closed) which will (hopefully) make its way to the develop branch in a few minutes.

@gong1 I can also add a comment on this if you like. Just to avoid later confusion :-)

@langguth1 cool! Free to add comment there!