Add dataset_kwargs in FineTuningDataModule

[jiajunly]

Is your feature request related to a problem? Please describe.

When executing prepare_data in FineTuningDataModule, the dataset_kwargs parameter is not passed. We may need it later when tokenize_dataset.

Describe the solution you'd like
I suggest adding this parameter.

        if not self.train_path_packed.is_file():
              prepare_packed_sequence_data(
                  input_path=self.train_path,
                  output_path=self.train_path_packed,
                  packed_sequence_size=self.packed_sequence_size,
                  tokenizer=self.tokenizer,
                  max_seq_length=self.seq_length,
                  seed=self.seed,
                  output_metadata_path=self.pack_metadata,
                  dataset_kwargs=self.dataset_kwargs, # here
              )

          if not self.validation_path_packed.is_file():
              prepare_packed_sequence_data(
                  input_path=self.validation_path,
                  output_path=self.validation_path_packed,
                  packed_sequence_size=self.packed_sequence_size,
                  tokenizer=self.tokenizer,
                  max_seq_length=self.seq_length,
                  seed=self.seed,
                  output_metadata_path=self.pack_metadata,
                  dataset_kwargs=self.dataset_kwargs, # here
              )

Describe alternatives you've considered

No

Additional context

here

[cuichenx]

Thanks for the suggestion! We have a WIP PR that achieves a similar purpose. Let me coordinate between you and the other author