Add paper link, task categories, and description to dataset card

README.md

@@ -1,4 +1,9 @@
 ---
+task_categories:
+- text-generation
+tags:
+- medical
+- ddxplus
 dataset_info:
   features:
   - name: problem
@@ -37,3 +42,38 @@ configs:
   - split: test
     path: data/test-*
 ---
+
+# medDataset_25k
+
+This dataset was introduced in the paper [Reaching Beyond the Mode: RL for Distributional Reasoning in Language Models](https://huggingface.co/papers/2603.24844).
+
+[**Project Page**](https://multi-answer-rl.github.io/) | [**GitHub Repository**](https://github.com/ishapuri/multi_answer_rl)
+
+## Dataset Summary
+
+`medDataset_25k` contains 25,000 patient cases derived from the DDXPlus dataset. It is designed to train and evaluate language models on distributional reasoning—the ability to generate multiple plausible hypotheses with associated confidence estimates, rather than collapsing to a single modal answer.
+
+Each sample in the dataset represents a clinical case consisting of a patient's clinical presentation and a ranked differential diagnosis.
+
+## Dataset Structure
+
+The dataset contains the following features:
+
+- `problem`: The clinical presentation of the patient (symptoms and history).
+- `Differential_Diagnoses`: A list of potential diagnoses.
+- `Differential_Diagnoses_ICD10s`: A list of ICD-10 codes corresponding to the diagnoses.
+- `answers`: The ground truth diagnostic answers.
+- `Differential_Probabilities`: Floating-point probabilities for each diagnosis in the differential.
+- `Differential_Answer_Text`: Textual representation of the differential diagnosis.
+- `source`: The origin of the data sample.
+
+## Citation
+
+```bibtex
+@article{puri2024reaching,
+  title={Reaching Beyond the Mode: RL for Distributional Reasoning in Language Models},
+  author={Puri, Isha and Damani, Mehul and Shenfeld, Idan and Ghassemi, Marzyeh and Andreas, Jacob and Kim, Yoon},
+  journal={arXiv preprint arXiv:2412.14642},
+  year={2024}
+}
+```