language:
- ms
- en
Dataset Summary
This dataset comprises of both the original English and Malay translated version of the FIB benchmark dataset. Each example in the original dataset consists of a document and a pair of summaries: a factually consistent one and a factually inconsistent one. It is based on documents and summaries from XSum and CNN/DM.
For our use case, we've exploded the rows in the dataset such that passage-summary pairs and a answer column that determines whether a the summary is factually consistent (1) or not (0).
Malay Version Generation
The Malay section of the dataset was done by translating the above dataset using OpenAI's 4o-mini. The notebook for this generating this dataset can be found here: https://github.com/wanadzhar913/aitinkerers-hackathon-supa-team-werecooked/blob/master/notebooks-data-preparation/01_dataset_prep_fib_openai.ipynb