Add link to paper, code repository, task category, and dataset description

This PR adds a link to the paper and the code repository for this dataset. It also sets the appropriate task category and adds a description of the dataset.

README.md

@@ -22,4 +22,12 @@ configs:
   data_files:
   - split: train
     path: data/train-*
+task_categories:
+- text-generation
 ---
+
+This dataset is used in the paper [Token Cleaning: Fine-Grained Data Selection for LLM Supervised Fine-Tuning](https://huggingface.co/papers/2502.01968).
+
+It is a collection of 50k samples constructed based on a new powerful data curation pipeline proposed by [DS2](https://openreview.net/pdf?id=DKkQtRMowq), which involves selecting data samples using quality rating scores generated by LLMs.
+
+Code: https://github.com/JlPang863/LLM_token_selection