Update README.md
Browse files
README.md
CHANGED
|
@@ -1757,5 +1757,6 @@ configs:
|
|
| 1757 |
tags:
|
| 1758 |
- sports
|
| 1759 |
---
|
|
|
|
| 1760 |
|
| 1761 |
OnlySports Dataset, a comprehensive collection of English sports documents. This dataset comprises a diverse range of content including news articles, blogs, match reports, interviews, and tutorials, all extracted from the FineWeb dataset. FineWeb is a thoroughly cleaned and deduplicated subset of CommonCrawl, spanning from 2013 to present. It represents one of the best open-source datasets for LLM training. Our extraction process involved two key steps: first, we applied URL filtering to identify potentially relevant content, and second, we developed a custom sports text classifier to accurately identify and extract sports-related documents from the filtered data. The resulting OnlySports Dataset encompasses 1.2 TB of text, equivalent to approximately 600 billion RWKV/GPT2 tokens. This makes it the largest sport domain dataset to date, significantly surpassing previous collections in both scale and comprehensiveness
|
|
|
|
| 1757 |
tags:
|
| 1758 |
- sports
|
| 1759 |
---
|
| 1760 |
+
Readme will be updated shortly, for now use https://github.com/chrischenhub/OnlySportsLM for reference.
|
| 1761 |
|
| 1762 |
OnlySports Dataset, a comprehensive collection of English sports documents. This dataset comprises a diverse range of content including news articles, blogs, match reports, interviews, and tutorials, all extracted from the FineWeb dataset. FineWeb is a thoroughly cleaned and deduplicated subset of CommonCrawl, spanning from 2013 to present. It represents one of the best open-source datasets for LLM training. Our extraction process involved two key steps: first, we applied URL filtering to identify potentially relevant content, and second, we developed a custom sports text classifier to accurately identify and extract sports-related documents from the filtered data. The resulting OnlySports Dataset encompasses 1.2 TB of text, equivalent to approximately 600 billion RWKV/GPT2 tokens. This makes it the largest sport domain dataset to date, significantly surpassing previous collections in both scale and comprehensiveness
|