Datasets:

Modalities:
Text
Formats:
parquet
Languages:
English
Libraries:
Datasets
Dask
License:
pretrain_en / README.md
i4never's picture
Update README.md
916f804
---
dataset_info:
features:
- name: content
dtype: string
splits:
- name: train
num_bytes: 48490123196
num_examples: 22690306
download_size: 5070161762
dataset_size: 48490123196
license: apache-2.0
task_categories:
- text-generation
language:
- en
size_categories:
- 10M<n<100M
---
# Dataset Card for "pretrain_en"
[Tigerbot](https://github.com/TigerResearch/TigerBot) pretrain数据的英文部分。
## Usage
```python
import datasets
ds_sft = datasets.load_dataset('TigerResearch/pretrain_en')
```