metadata
license: mit
task_categories:
- text-generation
configs:
- config_name: default
data_files:
- split: train
path: data/train_*
- split: test
path: data/test_*
We collect a 2.5B training dataset from various domains for long-context continual pre-training. The composition of this dataset is as follows (partially inspired by Long-Data-Collection):
| Domain | Proportion | Source |
|---|---|---|
| Book | 40% | Redpajama-Book |
| Arxiv | 20% | Redpajama-Arxiv |
| General | 20% | Redpajama |
| Code | 10% | LCC-Python |
| QA | 5% | Natural Questions |
| Summarization | 5% | BookSum |
We have also curated a test dataset comprising 250 million tokens, mirroring the same composition. The selection criteria ensured that the average n-gram similarity (for n=2, 3, 4) with the training set is below 10%. This threshold effectively excludes all QA and Summarization data, resulting in a test corpus where the distribution of tokens across Book, Arxiv, General, and Code categories follows a ratio of 4:2:2:1, respectively.