Improve dataset card with paper information and task category

This PR improves the dataset card by:

- Adding a link to the paper: [From Hours to Minutes: Lossless Acceleration of Ultra Long Sequence Generation up to 100K Tokens](https://hf.co/papers/2502.18890)
- Adding a link to the Github repository: https://github.com/bigai-nlco/TokenSwift
- Specifying the `task_categories` as `text-generation`
- Adding relevant tags.
- Adding license information (assuming MIT license based on common open-source practices, please adjust if incorrect)

README.md

@@ -1,4 +1,11 @@
 ---
+license: mit
+task_categories:
+  - text-generation
+tags:
+  - long-sequence-generation
+  - training-data
+  - llm-training
 dataset_info:
   features:
   - name: input_ids
@@ -15,3 +22,9 @@ configs:
   - split: train
     path: data/train-*
 ---
+
+This dataset provides training data for accelerating ultra-long sequence generation with large language models (LLMs).  It's used in the paper [From Hours to Minutes: Lossless Acceleration of Ultra Long Sequence Generation up to 100K Tokens](https://hf.co/papers/2502.18890).
+
+The data is derived from the PG-19 dataset, filtering out sequences smaller than 8K tokens to focus on ultra-long sequences relevant for the TokenSwift method described in the paper.
+
+Code: https://github.com/bigai-nlco/TokenSwift