---
license: apache-2.0
language:
- en
task_categories:
- text-classification
- text-generation
tags:
- Stigma
- Clinical
- LLMs
pretty_name: >-
  Data for paper "Large Language Models Generate Stigmatizing Language During
  Reasoning Over Real-World Clinical Data"
size_categories:
- 10M<n<100M
---

# LLM Stigma Data
This dataset contains the data for the paper "[Large Language Models Generate Stigmatizing Language During Reasoning Over Real-World Clinical Data](https://www.medrxiv.org/content/10.64898/2026.08.12.26360210v1)". It contains the system prompt, user input, true labels, and LLM outputs of the main and the destigmatizing experiment. The data is packed on the model level. Each tar.gz file contains 13 JSON files.

**Note: Only the tasks that are allowed to be open-sourced (N = 13) are included. For the rest of tasks (N = 22), please contact the authors for access.**

Each entry of the JSON file has the following field names:
- task: Name of the task
- type: Task type ("clf" means classification, "ext" means extraction, "gen" means generation)
- language: Task language (e.g., "en" means that this is an English task)
- id: ID of the case (The ids shown start with the first id of the test set.)
- split: Set split ("test" means the test set. Only the test set is used in this study.)
- instruction: The system prompt
- input: The user input
- output: The true label
- pred: The model's output