---
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](https://huggingface.co/datasets/togethercomputer/Long-Data-Collections)):

| Domain        | Proportion | Source |
| ------------- | ---------- | ------ |
| Book          | 40%        | [Redpajama-Book](https://huggingface.co/datasets/togethercomputer/RedPajama-Data-1T)   |
| Arxiv         | 20%        | [Redpajama-Arxiv](https://huggingface.co/datasets/togethercomputer/RedPajama-Data-1T)    |
| General       | 20%        | [Redpajama](https://huggingface.co/datasets/togethercomputer/RedPajama-Data-1T)    |
| Code          | 10%        | [LCC-Python](https://huggingface.co/datasets/microsoft/LCC_python)    |
| QA            | 5%         | [Natural Questions](https://ai.google.com/research/NaturalQuestions/)   |
| Summarization | 5%         | [BookSum](https://github.com/salesforce/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.