Keiran Paster
commited on
Commit
•
7e35f5a
1
Parent(s):
d01f53f
add figures
Browse files- README.md +32 -1
- imgs/contents_pie_chart.png +3 -0
README.md
CHANGED
|
@@ -19,4 +19,35 @@ dataset_info:
|
|
| 19 |
# <img align="left" width="50" height="50" src="imgs/openwebmath_logo.png" style="margin-top: -12px; margin-bottom: 0px;"> OpenWebMath
|
| 20 |
|
| 21 |
|
| 22 |
-
**OpenWebMath** is a dataset containing the majority of the high-quality, mathematical text from the internet. It is filtered and extracted from over 200B HTML files on Common Crawl down to a set of 6.3 million documents containing 14.7B tokens.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 19 |
# <img align="left" width="50" height="50" src="imgs/openwebmath_logo.png" style="margin-top: -12px; margin-bottom: 0px;"> OpenWebMath
|
| 20 |
|
| 21 |
|
| 22 |
+
**OpenWebMath** is a dataset containing the majority of the high-quality, mathematical text from the internet. It is filtered and extracted from over 200B HTML files on Common Crawl down to a set of 6.3 million documents containing 14.7B tokens. OpenWebMath is intended for use in *pretraining and finetuning large language models*.
|
| 23 |
+
|
| 24 |
+
## Contents
|
| 25 |
+
|
| 26 |
+
Our aim with OpenWebMath was to build a dataset of as many mathematical documents sourced from the web as possible while preserving the formatting of mathematical content such as LaTeX equations. OpenWebMath contains documents from over 130k different domains, ranging from forums, educational and reference content, and blogs. The dataset contains documents covering mathematics, physics, statistics, computer science, and more.
|
| 27 |
+
|
| 28 |
+
**Table 1:** Most Common Domains in OpenWebMath by Character Count
|
| 29 |
+
| Domain | # Characters | % Characters |
|
| 30 |
+
|-----------------------|--------------|--------------|
|
| 31 |
+
| stackexchange.com | 4,655,132,784| 9.55% |
|
| 32 |
+
| nature.com | 1,529,935,838| 3.14% |
|
| 33 |
+
| wordpress.com | 1,294,166,938| 2.66% |
|
| 34 |
+
| physicsforums.com | 1,160,137,919| 2.38% |
|
| 35 |
+
| github.io | 725,689,722 | 1.49% |
|
| 36 |
+
| zbmath.org | 620,019,503 | 1.27% |
|
| 37 |
+
| wikipedia.org | 618,024,754 | 1.27% |
|
| 38 |
+
| groundai.com | 545,214,990 | 1.12% |
|
| 39 |
+
| blogspot.com | 520,392,333 | 1.07% |
|
| 40 |
+
| mathoverflow.net | 499,102,560 | 1.02% |
|
| 41 |
+
| gmatclub.com | 442,611,169 | 0.91% |
|
| 42 |
+
| gamedev.net | 426,478,461 | 0.88% |
|
| 43 |
+
| ac.uk | 402,111,665 | 0.83% |
|
| 44 |
+
| aimsciences.org | 344,716,386 | 0.71% |
|
| 45 |
+
| mathhelpforum.com | 319,215,756 | 0.65% |
|
| 46 |
+
| deepai.org | 313,512,520 | 0.64% |
|
| 47 |
+
| libretexts.org | 282,014,149 | 0.58% |
|
| 48 |
+
| readthedocs.io | 269,816,413 | 0.55% |
|
| 49 |
+
| tib.eu | 199,714,017 | 0.41% |
|
| 50 |
+
| mit.edu | 198,487,362 | 0.41% |
|
| 51 |
+
|
| 52 |
+
**Figure 1:** Content Distribution of OpenWebMath by Type and Subject
|
| 53 |
+
|
imgs/contents_pie_chart.png
ADDED
|
Git LFS Details
|