GitLab

Created a link between the linear layer bias and the model attribute bias. This does not change anything for the user nor for the conversion scripts, but allows the `resize_token_embeddings` method to resize the bias as well as the weights of the decoder.

Added a test.

Releasing file lock

Update pipelines.py

* Added repetition penalty

* Default PPLM repetition_penalty to neutral

* Minor modifications to comply with reviewer's suggestions. (j -> token_idx)

* Formatted code with `make style`

T5: move rp_bucket to relative_attention_bias' device

otherwise, `rp_bucket` will always be on cpu and fail if `self.relative_attention_bias` is on cuda