Include link to Nomic Atlas map in Dataset Card
#16
by
rguo123
- opened
README.md
CHANGED
@@ -67,6 +67,13 @@ Falcon RefinedWeb was created to serve as an English large-scale dataset for the
|
|
67 |
|
68 |
It was built on top of CommonCrawl, leveraging stringent filtering and extensive deduplication.
|
69 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
70 |
### Supported Tasks and Leaderboards
|
71 |
|
72 |
RefinedWeb is intended to be primarly used as a pretraining dataset for large language models. Practitioners may leverage it for upstream evaluation with a validation loss, but we do not provide any canonical split.
|
|
|
67 |
|
68 |
It was built on top of CommonCrawl, leveraging stringent filtering and extensive deduplication.
|
69 |
|
70 |
+
### Dataset Visualization
|
71 |
+
Click the [Nomic Atlas](https://atlas.nomic.ai/map/refined_web) map below to visualize a 10 million subsample of RefinedWeb.
|
72 |
+
|
73 |
+
<a href="https://atlas.nomic.ai/map/refined_web">
|
74 |
+
<img src="https://cdn-uploads.huggingface.co/production/uploads/6480c476cacb1c4a0696eeb8/HKNKUloDBH9xVcj6MJGv4.webp" alt="Nomic-Atlas RefinedWeb Map" width="30%"/>
|
75 |
+
</a>
|
76 |
+
|
77 |
### Supported Tasks and Leaderboards
|
78 |
|
79 |
RefinedWeb is intended to be primarly used as a pretraining dataset for large language models. Practitioners may leverage it for upstream evaluation with a validation loss, but we do not provide any canonical split.
|