Datasets:

tamarsonha
/

MUSE-News-Train

dataset_info:
  features:
    - name: text
      dtype: string
  splits:
    - name: full
      num_bytes: 26113818
      num_examples: 7109
    - name: retain
      num_bytes: 12894207
      num_examples: 3555
  download_size: 23031717
  dataset_size: 39008025
configs:
  - config_name: default
    data_files:
      - split: full
        path: data/full-*
      - split: retain
        path: data/retain-*

MUSE-News-Train

This dataset is a simple merger of the pretraining data from the original MUSE-News dataset.

Dataset Details

Dataset Sources [optional]

Repository: https://huggingface.co/datasets/muse-bench/MUSE-News
Paper: https://arxiv.org/pdf/2407.06460

Dataset Creation

To create this dataset, we simply started from the muse-bench dataset and selected the train subset. Then, by merging the retain1 and retain2 splits we get the actual retain subset, and by further merging this with the original forget split, we get the full dataset used for pre-training on the specific MUSE task.