---
configs:
- config_name: default
  data_files:
  - split: pubmed
    path: data/pubmed-*
  - split: nonbiomedical
    path: data/nonbiomedical-*
  - split: counterfactual
    path: data/counterfactual-*
  - split: casual
    path: data/casual-*
  - split: rap
    path: data/rap-*
dataset_info:
  features:
  - name: PubmedData
    struct:
    - name: ArticleIdList
      sequence:
      - name: ArticleId
        sequence: string
    - name: PublicationStatus
      dtype: string
    - name: History
      struct:
      - name: PubMedPubDate
        sequence:
        - name: Year
          dtype: int32
        - name: Month
          dtype: int32
        - name: Day
          dtype: int32
    - name: ReferenceList
      sequence:
      - name: Citation
        dtype: string
      - name: CitationId
        dtype: int32
  - name: text
    dtype: string
  splits:
  - name: pubmed
    num_bytes: 1166668
    num_examples: 1000
  - name: nonbiomedical
    num_bytes: 1141909
    num_examples: 1000
  - name: counterfactual
    num_bytes: 1179347
    num_examples: 991
  - name: casual
    num_bytes: 1205949
    num_examples: 1000
  - name: rap
    num_bytes: 1252260
    num_examples: 1000
  download_size: 3357032
  dataset_size: 5946133
language:
- en
---
# A corpus of rewritten pubmed abstracts

This corpus contains a 1k example subset from the [pubmed](https://huggingface.co/datasets/pubmed) corpus and various rewritten versions. The rewritten versions change one aspect of the orginal text and keeps other aspects unchanged as much as possible.

- **Paper:** [Dissecting learning and forgetting in language model finetuning](https://openreview.net/forum?id=tmsqb6WpLz)

Another corpus of rewritten general text is provided here: [c4_derived](https://huggingface.co/datasets/xiaozeroone/c4_derived)


### Data Splits

- pubmed: a 1k example subset from the original pubmed corpus

- nonbiomedical: main topic of text changed to nonbiomedical topic

- counerfactual: factuals knowledge in text replaced by incorrect factuals

- casual: style of text changed to a casual style

- rap: style of text changed to a rap style

## Dataset Creation

Text is generated by ChatGPT with corresponding prompts. Refer to the paper for the instructions used to generate text in each derived subsets.

Please check the terms and conditions of pubmed data [here](https://www.nlm.nih.gov/databases/download/terms_and_conditions.html).


### Citation Information

```
@inproceedings{
  zhang2024dissecting,
  title={Dissecting learning and forgetting in language model finetuning},
  author={Xiao Zhang and Ji Wu},
  booktitle={The Twelfth International Conference on Learning Representations},
  year={2024},
  url={https://openreview.net/forum?id=tmsqb6WpLz}
}
```