GitLab

The introduction of a decoder introduces 2 changes:
- We need to be able to specify a separate mask in the cross
attention to mask the positions corresponding to padding tokens in the
encoder state.
- The self-attention in the decoder needs to be causal on top of not
attending to padding tokens.

the definition of `get_masks` would blow with the proper combination of
arguments. It was just a matter of moving a definition outside of a
control structure.

We currenctly instantiate encoders and decoders for the seq2seq by
passing the `is_decoder` keyword argument to the `from_pretrained`
classmethod. On the other hand, the model class looks for the value
of the `is_decoder` attribute in its config.

In order for the value to propagate from the kwarg to the configuration
we simply need to define `is_decoder` as an attribute to the base
`PretrainedConfig`, with a default at `False`.

the data provided by Li Dong et al. were already tokenized, which means
that they are not compatible with  all the models in the library. We
thus process the raw data directly and tokenize them using the models'
tokenizers.

We write a function to load an preprocess the CNN/Daily Mail dataset as
provided by Li Dong et al. The issue is that this dataset has already
been tokenized by the authors, so we actually need to find the original,
plain-text dataset if we want to apply it to all models.

In Rothe et al.'s "Leveraging Pre-trained Checkpoints for Sequence
Generation Tasks", Bert2Bert is initialized with pre-trained weights for
the encoder, and only pre-trained embeddings for the decoder. The
current version of the code completely randomizes the weights of the
decoder.

We write a custom function to initiliaze the weights of the decoder; we
first initialize the decoder with the weights and then randomize
everything but the embeddings.

Since the preloading of weights relies on the name of the class's
attributes changing the namespace breaks loading pretrained weights on
Bert and all related models. I reverted `self_attention` to `attention`
and us `crossattention` for the decoder instead.

In the seq2seq model we need to both load pretrained weights in the
encoder and initialize the decoder randomly. Because the
`from_pretrained` method defined in the base class relies on module
names to assign weights, it would also initialize the decoder with
pretrained weights. To avoid this we override the method to only
initialize the encoder with pretrained weights.

I am not sure what happens when the class is initialized with the
pretrained weights.