---
license: mit
---

See our [Example.ipynb](./Example.ipynb)

## Model Overview

Trained and evaluated on mixed (noun-phrase and claim) targets from subtask A of EZStance:
```
@inproceedings{zhao-caragea-2024-ez,
    title = "{EZ}-{STANCE}: A Large Dataset for {E}nglish Zero-Shot Stance Detection",
    author = "Zhao, Chenye  and
      Caragea, Cornelia",
    editor = "Ku, Lun-Wei  and
      Martins, Andre  and
      Srikumar, Vivek",
    booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)",
    month = aug,
    year = "2024",
    address = "Bangkok, Thailand",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2024.acl-long.838/",
    doi = "10.18653/v1/2024.acl-long.838",
    pages = "15697--15714",
}
```

Used the `BART-MNLI-e` architecture from the same paper.
Weights were initialized from [facebook/bart-large-mnli](https://huggingface.co/facebook/bart-large-mnli):

Obtained macro F1-score of 0.82 (see [exp\_results/metrics.csv](exp_results/metrics.csv)) on the test data.


## Dependencies

- `python>=3.9.22`
- `transformers>=4.51.0`
- `accelerate>=0.26.0`
- `torch>=2.7.0`