README.md

---
dataset_info:
  features:
  - name: question_id
    dtype: string
  - name: question
    dtype: string
  - name: options
    struct:
    - name: A
      dtype: string
    - name: B
      dtype: string
    - name: C
      dtype: string
    - name: D
      dtype: string
    - name: E
      dtype: string
    - name: F
      dtype: string
    - name: G
      dtype: string
  - name: answer
    dtype: string
  - name: src
    dtype: string
  splits:
  - name: test
    num_bytes: 109817
    num_examples: 100
  download_size: 49322
  dataset_size: 109817
configs:
- config_name: default
  data_files:
  - split: test
    path: data/test-*
license: apache-2.0
---

# M-ARC

HuggingFace upload of a clinical QA benchmark designed to exploit LLMs' "inductive biases toward inflexible pattern matching from their training data rather than
engaging in flexible reasoning." If used, please cite the original authors using the citation below.

## Dataset Details

### Dataset Description

The dataset contains one split:
  - **test**: up to seven-option multiple-choice QA (choices A-G)

### Dataset Sources

- **Repository:** https://github.com/dbernardo05/medARC-QA
- **Paper:** https://arxiv.org/pdf/2502.04381

### Direct Use

```python
import json
from datasets import load_dataset

if __name__ == "__main__":
    # load the test split
    dataset_test = load_dataset("mkieffer/M-ARC", split="test")
    print("\test split:\n", dataset_test)
    print("\ntest sample:\n", json.dumps(dataset_test[0], indent=2))
```


## Citation 

```
@misc{kim2025limitationslargelanguagemodels,
      title={Limitations of Large Language Models in Clinical Problem-Solving Arising from Inflexible Reasoning}, 
      author={Jonathan Kim and Anna Podlasek and Kie Shidara and Feng Liu and Ahmed Alaa and Danilo Bernardo},
      year={2025},
      eprint={2502.04381},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2502.04381}, 
}
```