Create README.md
Browse files
README.md
ADDED
|
@@ -0,0 +1,44 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
Dataset Summary
|
| 2 |
+
This dataset is a 2 Billion token subset of the RedPajama-Data-1T dataset created by Together Computer.
|
| 3 |
+
|
| 4 |
+
It was created to provide a smaller, lightweight version of the massive 1.2 Trillion token dataset for testing, debugging, and small-scale pre-training experiments. This specific subset was streamed primarily from the C4 (Colossal Clean Crawled Corpus) partition of the original RedPajama dataset.
|
| 5 |
+
|
| 6 |
+
Dataset Structure
|
| 7 |
+
The dataset retains the original structure of RedPajama. Each row contains:
|
| 8 |
+
text: The actual content (string).
|
| 9 |
+
meta: Metadata dict containing url, timestamp, source, etc.
|
| 10 |
+
subset: The name of the RedPajama subset this sample came from (e.g., c4).
|
| 11 |
+
|
| 12 |
+
Source Data
|
| 13 |
+
This dataset is a direct downstream derivative of RedPajama-Data-1T.
|
| 14 |
+
|
| 15 |
+
Original Repository: togethercomputer/RedPajama-Data-1T
|
| 16 |
+
|
| 17 |
+
Original Author: Together Computer
|
| 18 |
+
|
| 19 |
+
Subset Used: c4 (Colossal Clean Crawled Corpus)
|
| 20 |
+
|
| 21 |
+
Creation Process
|
| 22 |
+
This dataset was generated using the Hugging Face datasets library with the following methodology:
|
| 23 |
+
|
| 24 |
+
Direct Streaming: Data was streamed directly from the original source files to avoid downloading the full 5TB dataset.
|
| 25 |
+
|
| 26 |
+
Selection: The first ~2 Billion tokens were selected from the c4 split.
|
| 27 |
+
|
| 28 |
+
Token Counting: Token counts were estimated using the EleutherAI/gpt-neox-20b tokenizer (or a 1 token ≈ 4 characters approximation).
|
| 29 |
+
|
| 30 |
+
Licensing Information
|
| 31 |
+
Since this is a subset of RedPajama, it inherits the licensing terms of the original data.
|
| 32 |
+
RedPajama-Data-1T is distributed under the licenses of its underlying data sources.
|
| 33 |
+
C4 (Common Crawl): Terms of Use are available here.
|
| 34 |
+
Code/Scripts: The code used to generate this subset is Open Source.
|
| 35 |
+
|
| 36 |
+
Citation
|
| 37 |
+
If you use this dataset, please cite the original RedPajama paper:
|
| 38 |
+
@software{together2023redpajama,
|
| 39 |
+
author = {Together Computer},
|
| 40 |
+
title = {RedPajama: An Open Source Recipe to Reproduce LLaMA training dataset},
|
| 41 |
+
month = April,
|
| 42 |
+
year = 2023,
|
| 43 |
+
url = {https://github.com/togethercomputer/RedPajama-Data}
|
| 44 |
+
}
|