factual-consistency-t5-large

Original Repo

Paper

Inference script available here

Data:

• Training Data
• Evaluation Benchmark

Results

Comparison of different factuality evaluation methods across all test datasets. The methods are ranked based on the Mean Win Rate, which measures overall performance on factuality tasks. The Average AUC column represents the average of all individual AUC-ROC scores.

Cite this work as follows:

@misc{agarwal2024zeroshotfactualconsistencyevaluation,
      title={Zero-shot Factual Consistency Evaluation Across Domains}, 
      author={Raunak Agarwal},
      year={2024},
      eprint={2408.04114},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2408.04114}, 
}