Improve dataset card: Add paper/code links, tasks, tags, description, sample usage, and citation

README.md

@@ -1,4 +1,66 @@
 ---
 license: apache-2.0
+task_categories:
+- text-generation
+- reinforcement-learning
+language:
+- en
+tags:
+- llm
+- math
+- reasoning
+- fine-tuning
 ---
-Selected easy prompts used to train Qwen2.5-Math-7B.
+
+This dataset contains selected easy prompts used to train Qwen2.5-Math-7B, as part of the research presented in the paper [Reinforce-Ada: An Adaptive Sampling Framework for Reinforce-Style LLM Training](https://huggingface.co/papers/2510.04996).
+
+Reinforce-Ada is an adaptive sampling framework for online reinforcement learning (RL) post-training of large language models (LLMs) for reasoning tasks. It aims to resolve the "signal collapse" problem by continuously reallocating sampling effort to prompts with the greatest uncertainty or learning potential. This dataset provides specific prompts utilized in the experiments to facilitate this adaptive sampling process.
+
+**Paper:** [https://huggingface.co/papers/2510.04996](https://huggingface.co/papers/2510.04996)
+**Code:** [https://github.com/RLHFlow/Reinforce-Ada](https://github.com/RLHFlow/Reinforce-Ada)
+
+## Sample Usage
+
+To prepare and process the training data for the Reinforce-Ada framework, you can use the scripts provided in the associated GitHub repository.
+
+First, prepare the training and test datasets. You can adjust the `pass_rate` for hard and easy prompt selection:
+
+```bash
+# adjust pass_rate to 0.125 and 0.313 for hard and easy prompt selection, respectively.
+bash scripts/prepare_data.py
+```
+
+After preparing the data, convert it to the `verl` training format and generate a validation set using the following commands:
+
+```bash
+# Convert to verl training format
+echo "Converting to verl training format..."
+python3 data_process/reformat.py \
+    --local_dir ${output_dir} \
+    --model_name_or_path ${model_name} \
+    --data_source ${data_name} \
+
+# Generate validation set
+echo "Generating validation set..."
+python3 data_process/get_validation_set.py \
+    --local_dir ${output_dir} \
+    --model_name_or_path ${model_name}
+```
+
+For more details on environment setup, experimentation, and using the processed training sets and checkpoints, please refer to the [Reinforce-Ada GitHub repository](https://github.com/RLHFlow/Reinforce-Ada).
+
+## Citation
+
+If you find our paper or code helpful, please cite our work:
+
+```bibtex
+@misc{xiong2025reinforceada,
+      title={Reinforce-Ada: An Adaptive Sampling Framework for Reinforce-Style LLM Training}, 
+      author={Wei Xiong and Chenlu Ye and Baohao Liao and Hanze Dong and Xinxing Xu and Christof Monz and Jiang Bian and Nan Jiang and Tong Zhang},
+      year={2025},
+      eprint={2510.04996},
+      archivePrefix={arXiv},
+      primaryClass={cs.LG},
+      url={https://arxiv.org/abs/2510.04996}, 
+}
+```