---
license: mit
task_categories:
- question-answering
language:
- en
---
Credibility-aware Generation Benchmark (CAGB) is a benchmark constructed to evaluate the credibility-aware generation ability of models, dealing with flawed information in the context.
This benchmark encompasses the following three specific scenarios where the integration of credibility is essential:
- Open-domain QA
  - 2WikiMultiHopQA
  - HotpotQA
  - Musique
  - RGB
- Time-sensitive QA
  - EvoTempQA
- Misinformation Polluted QA
  - NewsPollutedQA

Read our [paper](https://arxiv.org/abs/2404.06809) for more insights on credibility-aware generation.