| dataset_info: | |
| features: | |
| - name: input_ids | |
| sequence: int32 | |
| splits: | |
| - name: train | |
| num_bytes: 1076886272 | |
| num_examples: 27332 | |
| download_size: 487651276 | |
| dataset_size: 1076886272 | |
| task_categories: | |
| - text-generation | |
| configs: | |
| - config_name: default | |
| data_files: | |
| - split: train | |
| path: data/train-* | |
| This dataset contains data used in the paper [From Hours to Minutes: Lossless Acceleration of Ultra Long Sequence Generation up to 100K Tokens](https://huggingface.co/papers/2502.18890). | |
| Code: https://github.com/bigai-nlco/TokenSwift |