Datasets:
Update README.md
Browse files
README.md
CHANGED
@@ -24,7 +24,7 @@ language:
|
|
24 |
- es
|
25 |
configs:
|
26 |
- config_name: v1
|
27 |
-
data_files:
|
28 |
- data/CC-MAIN-2019-30/**/*.parquet
|
29 |
- data/CC-MAIN-2020-05/**/*.parquet
|
30 |
- data/CC-MAIN-2022-05/**/*.parquet
|
@@ -34,7 +34,6 @@ configs:
|
|
34 |
- data/CC-MAIN-2025-05/**/*.parquet
|
35 |
- config_name: default
|
36 |
data_files: data/**/*.parquet
|
37 |
-
# Languages
|
38 |
- config_name: afr
|
39 |
data_files: data/**/afr/*.parquet
|
40 |
- config_name: deu
|
@@ -51,8 +50,6 @@ configs:
|
|
51 |
data_files: data/**/ita/*.parquet
|
52 |
- config_name: nld
|
53 |
data_files: data/**/nld/*.parquet
|
54 |
-
# Per-crawl
|
55 |
-
# CC-MAIN-2019-30
|
56 |
- config_name: CC-MAIN-2019-30
|
57 |
data_files: data/CC-MAIN-2019-30/**/*.parquet
|
58 |
- config_name: CC-MAIN-2019-30-afr
|
@@ -71,7 +68,6 @@ configs:
|
|
71 |
data_files: data/CC-MAIN-2019-30/ita/*.parquet
|
72 |
- config_name: CC-MAIN-2019-30-nld
|
73 |
data_files: data/CC-MAIN-2019-30/nld/*.parquet
|
74 |
-
# CC-MAIN-2020-05
|
75 |
- config_name: CC-MAIN-2020-05
|
76 |
data_files: data/CC-MAIN-2020-05/**/*.parquet
|
77 |
- config_name: CC-MAIN-2020-05-afr
|
@@ -90,7 +86,6 @@ configs:
|
|
90 |
data_files: data/CC-MAIN-2020-05/ita/*.parquet
|
91 |
- config_name: CC-MAIN-2020-05-nld
|
92 |
data_files: data/CC-MAIN-2020-05/nld/*.parquet
|
93 |
-
# CC-MAIN-2022-05
|
94 |
- config_name: CC-MAIN-2022-05
|
95 |
data_files: data/CC-MAIN-2022-05/**/*.parquet
|
96 |
- config_name: CC-MAIN-2022-05-afr
|
@@ -109,7 +104,6 @@ configs:
|
|
109 |
data_files: data/CC-MAIN-2022-05/ita/*.parquet
|
110 |
- config_name: CC-MAIN-2022-05-nld
|
111 |
data_files: data/CC-MAIN-2022-05/nld/*.parquet
|
112 |
-
# CC-MAIN-2023-06
|
113 |
- config_name: CC-MAIN-2023-06
|
114 |
data_files: data/CC-MAIN-2023-06/**/*.parquet
|
115 |
- config_name: CC-MAIN-2023-06-afr
|
@@ -128,7 +122,6 @@ configs:
|
|
128 |
data_files: data/CC-MAIN-2023-06/ita/*.parquet
|
129 |
- config_name: CC-MAIN-2023-06-nld
|
130 |
data_files: data/CC-MAIN-2023-06/nld/*.parquet
|
131 |
-
# CC-MAIN-2024-46
|
132 |
- config_name: CC-MAIN-2024-46
|
133 |
data_files: data/CC-MAIN-2024-46/**/*.parquet
|
134 |
- config_name: CC-MAIN-2024-46-afr
|
@@ -147,7 +140,6 @@ configs:
|
|
147 |
data_files: data/CC-MAIN-2024-46/ita/*.parquet
|
148 |
- config_name: CC-MAIN-2024-46-nld
|
149 |
data_files: data/CC-MAIN-2024-46/nld/*.parquet
|
150 |
-
# CC-MAIN-2024-51
|
151 |
- config_name: CC-MAIN-2024-51
|
152 |
data_files: data/CC-MAIN-2024-51/**/*.parquet
|
153 |
- config_name: CC-MAIN-2024-51-afr
|
@@ -166,7 +158,6 @@ configs:
|
|
166 |
data_files: data/CC-MAIN-2024-51/ita/*.parquet
|
167 |
- config_name: CC-MAIN-2024-51-nld
|
168 |
data_files: data/CC-MAIN-2024-51/nld/*.parquet
|
169 |
-
# CC-MAIN-2025-05
|
170 |
- config_name: CC-MAIN-2025-05
|
171 |
data_files: data/CC-MAIN-2025-05/**/*.parquet
|
172 |
- config_name: CC-MAIN-2025-05-afr
|
@@ -307,25 +298,11 @@ Based on these criteria, the "best" license is picked as the one in the `license
|
|
307 |
- Unsurpisingly, the data contains a lot of Wikipedia/Wikimedia content. Depending on what you need, you may wish to filter those out. For Wikipedia specifically, you may opt to use the more thoroughly parsed (but potentially more outdated) [wikimedia/wikipedia](https://huggingface.co/datasets/wikimedia/wikipedia) set.
|
308 |
- In exceptional cases, a link to creativecommons.org is found but the exact license could not be found. These are under `license_abbr="cc-unknown"` which you may wish to filter out.
|
309 |
|
|
|
310 |
|
311 |
-
|
|
|
312 |
|
313 |
-
```python
|
314 |
-
from datasets import load_dataset
|
315 |
-
|
316 |
-
|
317 |
-
ds = load_dataset("BramVanroy/CommonCrawl-CreativeCommons", "CC-MAIN-2019-30", split="train")
|
318 |
-
ds = ds.filter(
|
319 |
-
lambda x: (
|
320 |
-
(not x["license_disagreement"]) and # Only use pages with a consistent license
|
321 |
-
x["found_in_fw"] and # Only use pages that are in FineWeb(-2)
|
322 |
-
"nc" not in x["license_abbr"] and # Exclude non-commercial licenses
|
323 |
-
x["license_abbr"] != "cc-unknown" and # Exclude unknown licenses
|
324 |
-
"wiki" not in x["url"] # Exclude Wiki-like pages (best to get those from a more reliable parser)
|
325 |
-
),
|
326 |
-
num_proc=16
|
327 |
-
)
|
328 |
-
```
|
329 |
|
330 |
## Personal and Sensitive Information
|
331 |
|
@@ -370,4 +347,4 @@ If you use or modify [the software](https://github.com/BramVanroy/CommonCrawl-Cr
|
|
370 |
- The computational resources and services used in this work were provided by the [VSC (Flemish Supercomputer Center)](https://www.vscentrum.be/), funded by the Research Foundation Flanders (FWO) and the Flemish Government – department EWI under grant 2024-107.
|
371 |
- Guilherme Penedo ([@guipenedo](https://huggingface.co/guipenedo)) and the rest of the [FineWeb](https://huggingface.co/datasets/HuggingFaceFW/fineweb) and [datatrove](https://github.com/huggingface/datatrove) team for the help and insights
|
372 |
- [TNO](https://www.tno.nl/nl/), who funded the work hours to accomplish this code. They intend to use (parts of) [the generated material](https://huggingface.co/datasets/BramVanroy/CommonCrawl-CreativeCommons) for the [GPT-NL project](https://gpt-nl.nl/).
|
373 |
-
- ML6 and specifically Robin Van Craenenbroek for their [Fondant Creative Commons](https://github.com/ml6team/fondant-usecase-filter-creative-commons/tree/add-fondant-usecase-cc-image-extraction) filter for image datasets. While my approach is different, their code did serve as inspiration.
|
|
|
24 |
- es
|
25 |
configs:
|
26 |
- config_name: v1
|
27 |
+
data_files:
|
28 |
- data/CC-MAIN-2019-30/**/*.parquet
|
29 |
- data/CC-MAIN-2020-05/**/*.parquet
|
30 |
- data/CC-MAIN-2022-05/**/*.parquet
|
|
|
34 |
- data/CC-MAIN-2025-05/**/*.parquet
|
35 |
- config_name: default
|
36 |
data_files: data/**/*.parquet
|
|
|
37 |
- config_name: afr
|
38 |
data_files: data/**/afr/*.parquet
|
39 |
- config_name: deu
|
|
|
50 |
data_files: data/**/ita/*.parquet
|
51 |
- config_name: nld
|
52 |
data_files: data/**/nld/*.parquet
|
|
|
|
|
53 |
- config_name: CC-MAIN-2019-30
|
54 |
data_files: data/CC-MAIN-2019-30/**/*.parquet
|
55 |
- config_name: CC-MAIN-2019-30-afr
|
|
|
68 |
data_files: data/CC-MAIN-2019-30/ita/*.parquet
|
69 |
- config_name: CC-MAIN-2019-30-nld
|
70 |
data_files: data/CC-MAIN-2019-30/nld/*.parquet
|
|
|
71 |
- config_name: CC-MAIN-2020-05
|
72 |
data_files: data/CC-MAIN-2020-05/**/*.parquet
|
73 |
- config_name: CC-MAIN-2020-05-afr
|
|
|
86 |
data_files: data/CC-MAIN-2020-05/ita/*.parquet
|
87 |
- config_name: CC-MAIN-2020-05-nld
|
88 |
data_files: data/CC-MAIN-2020-05/nld/*.parquet
|
|
|
89 |
- config_name: CC-MAIN-2022-05
|
90 |
data_files: data/CC-MAIN-2022-05/**/*.parquet
|
91 |
- config_name: CC-MAIN-2022-05-afr
|
|
|
104 |
data_files: data/CC-MAIN-2022-05/ita/*.parquet
|
105 |
- config_name: CC-MAIN-2022-05-nld
|
106 |
data_files: data/CC-MAIN-2022-05/nld/*.parquet
|
|
|
107 |
- config_name: CC-MAIN-2023-06
|
108 |
data_files: data/CC-MAIN-2023-06/**/*.parquet
|
109 |
- config_name: CC-MAIN-2023-06-afr
|
|
|
122 |
data_files: data/CC-MAIN-2023-06/ita/*.parquet
|
123 |
- config_name: CC-MAIN-2023-06-nld
|
124 |
data_files: data/CC-MAIN-2023-06/nld/*.parquet
|
|
|
125 |
- config_name: CC-MAIN-2024-46
|
126 |
data_files: data/CC-MAIN-2024-46/**/*.parquet
|
127 |
- config_name: CC-MAIN-2024-46-afr
|
|
|
140 |
data_files: data/CC-MAIN-2024-46/ita/*.parquet
|
141 |
- config_name: CC-MAIN-2024-46-nld
|
142 |
data_files: data/CC-MAIN-2024-46/nld/*.parquet
|
|
|
143 |
- config_name: CC-MAIN-2024-51
|
144 |
data_files: data/CC-MAIN-2024-51/**/*.parquet
|
145 |
- config_name: CC-MAIN-2024-51-afr
|
|
|
158 |
data_files: data/CC-MAIN-2024-51/ita/*.parquet
|
159 |
- config_name: CC-MAIN-2024-51-nld
|
160 |
data_files: data/CC-MAIN-2024-51/nld/*.parquet
|
|
|
161 |
- config_name: CC-MAIN-2025-05
|
162 |
data_files: data/CC-MAIN-2025-05/**/*.parquet
|
163 |
- config_name: CC-MAIN-2025-05-afr
|
|
|
298 |
- Unsurpisingly, the data contains a lot of Wikipedia/Wikimedia content. Depending on what you need, you may wish to filter those out. For Wikipedia specifically, you may opt to use the more thoroughly parsed (but potentially more outdated) [wikimedia/wikipedia](https://huggingface.co/datasets/wikimedia/wikipedia) set.
|
299 |
- In exceptional cases, a link to creativecommons.org is found but the exact license could not be found. These are under `license_abbr="cc-unknown"` which you may wish to filter out.
|
300 |
|
301 |
+
Based on these recommendations, two subsets are available:
|
302 |
|
303 |
+
- [BramVanroy/CommonCrawl-CreativeCommons-fine](https://huggingface.co/datasets/BramVanroy/CommonCrawl-CreativeCommons-fine): only retaining items containing samples that are also in FineWeb(-2)
|
304 |
+
- [BramVanroy/CommonCrawl-CreativeCommons-strict](https://huggingface.co/datasets/BramVanroy/CommonCrawl-CreativeCommons-strict): additional filtering that also removes Wikipedia and non-commercial data
|
305 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
306 |
|
307 |
## Personal and Sensitive Information
|
308 |
|
|
|
347 |
- The computational resources and services used in this work were provided by the [VSC (Flemish Supercomputer Center)](https://www.vscentrum.be/), funded by the Research Foundation Flanders (FWO) and the Flemish Government – department EWI under grant 2024-107.
|
348 |
- Guilherme Penedo ([@guipenedo](https://huggingface.co/guipenedo)) and the rest of the [FineWeb](https://huggingface.co/datasets/HuggingFaceFW/fineweb) and [datatrove](https://github.com/huggingface/datatrove) team for the help and insights
|
349 |
- [TNO](https://www.tno.nl/nl/), who funded the work hours to accomplish this code. They intend to use (parts of) [the generated material](https://huggingface.co/datasets/BramVanroy/CommonCrawl-CreativeCommons) for the [GPT-NL project](https://gpt-nl.nl/).
|
350 |
+
- ML6 and specifically Robin Van Craenenbroek for their [Fondant Creative Commons](https://github.com/ml6team/fondant-usecase-filter-creative-commons/tree/add-fondant-usecase-cc-image-extraction) filter for image datasets. While my approach is different, their code did serve as inspiration.
|