Improve dataset card with paper information and task category
#1
by
nielsr
HF Staff
- opened
README.md
CHANGED
|
@@ -1,4 +1,11 @@
|
|
| 1 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 2 |
dataset_info:
|
| 3 |
features:
|
| 4 |
- name: input_ids
|
|
@@ -15,3 +22,9 @@ configs:
|
|
| 15 |
- split: train
|
| 16 |
path: data/train-*
|
| 17 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
---
|
| 2 |
+
license: mit
|
| 3 |
+
task_categories:
|
| 4 |
+
- text-generation
|
| 5 |
+
tags:
|
| 6 |
+
- long-sequence-generation
|
| 7 |
+
- training-data
|
| 8 |
+
- llm-training
|
| 9 |
dataset_info:
|
| 10 |
features:
|
| 11 |
- name: input_ids
|
|
|
|
| 22 |
- split: train
|
| 23 |
path: data/train-*
|
| 24 |
---
|
| 25 |
+
|
| 26 |
+
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).
|
| 27 |
+
|
| 28 |
+
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.
|
| 29 |
+
|
| 30 |
+
Code: https://github.com/bigai-nlco/TokenSwift
|