Publish shard CC-MAIN-2026-08/00333
Browse files- README.md +10 -10
- data/CC-MAIN-2026-08/00333.parquet +3 -0
- stats.csv +1 -0
README.md
CHANGED
|
@@ -32,7 +32,7 @@ configs:
|
|
| 32 |
|
| 33 |
Open Index is a large-scale web text dataset built from [Common Crawl](https://commoncrawl.org). Every page goes through a pipeline that extracts the main content from raw HTML, converts it to clean Markdown using [trafilatura](https://github.com/adbar/trafilatura), and packages the result into Parquet files with full WARC metadata preserved.
|
| 34 |
|
| 35 |
-
The dataset currently includes crawl **CC-MAIN-2026-08** with **15,
|
| 36 |
|
| 37 |
Open Index is released under the **Open Data Commons Attribution License (ODC-By) v1.0**, the same license used by Common Crawl.
|
| 38 |
|
|
@@ -164,27 +164,27 @@ Empty conversions (pages where trafilatura could not extract meaningful content)
|
|
| 164 |
|
| 165 |
### Compression Ratios
|
| 166 |
|
| 167 |
-
Numbers below are actual measurements summed across all
|
| 168 |
|
| 169 |
-
| Stage |
|
| 170 |
|---|---|---|---|
|
| 171 |
-
| Raw WARC (.warc.gz, downloaded) | ~
|
| 172 |
| HTML extracted (uncompressed) | 1.9 TB | ~295 TB | — |
|
| 173 |
-
| Packed markdown WARC (.md.warc.gz) | ~33.
|
| 174 |
| Final Parquet (Zstd level 19) | 22.5 GB | ~2.9 TB | **-32.4%** vs packed WARC |
|
| 175 |
|
| 176 |
-
The big win is the HTML → Markdown step: trafilatura strips all tags, scripts, styles, navigation, and ads, keeping only the main content. This cuts 1.9 TB of uncompressed HTML down to 70.
|
| 177 |
|
| 178 |
-
End to end: ~
|
| 179 |
|
| 180 |
### Processing Times
|
| 181 |
|
| 182 |
-
Pipeline timings across
|
| 183 |
|
| 184 |
```
|
| 185 |
Download (raw WARC) ████████████████████████ total 46h 34m 39s avg 3m 26s
|
| 186 |
-
Convert (HTML → MD) █░░░░░░░░░░░░░░░░░░░░░░░ total 3h
|
| 187 |
-
Export (Parquet) █████░░░░░░░░░░░░░░░░░░░ total 10h
|
| 188 |
Publish (HuggingFace) ███░░░░░░░░░░░░░░░░░░░░░ total 5h 59m 20s avg 26s
|
| 189 |
```
|
| 190 |
|
|
|
|
| 32 |
|
| 33 |
Open Index is a large-scale web text dataset built from [Common Crawl](https://commoncrawl.org). Every page goes through a pipeline that extracts the main content from raw HTML, converts it to clean Markdown using [trafilatura](https://github.com/adbar/trafilatura), and packages the result into Parquet files with full WARC metadata preserved.
|
| 34 |
|
| 35 |
+
The dataset currently includes crawl **CC-MAIN-2026-08** with **15,708,074 documents across 812 shards**. We plan to add more snapshots over time.
|
| 36 |
|
| 37 |
Open Index is released under the **Open Data Commons Attribution License (ODC-By) v1.0**, the same license used by Common Crawl.
|
| 38 |
|
|
|
|
| 164 |
|
| 165 |
### Compression Ratios
|
| 166 |
|
| 167 |
+
Numbers below are actual measurements summed across all 812 files of CC-MAIN-2026-08 (15,708,074 pages total), projected to the full crawl of 100,000 WARC files.
|
| 168 |
|
| 169 |
+
| Stage | 812 files (measured) | 100,000 files (projected) | Reduction |
|
| 170 |
|---|---|---|---|
|
| 171 |
+
| Raw WARC (.warc.gz, downloaded) | ~658.2 GB | ~83 TB | — |
|
| 172 |
| HTML extracted (uncompressed) | 1.9 TB | ~295 TB | — |
|
| 173 |
+
| Packed markdown WARC (.md.warc.gz) | ~33.3 GB | ~3.7 TB | **-98.3%** vs HTML |
|
| 174 |
| Final Parquet (Zstd level 19) | 22.5 GB | ~2.9 TB | **-32.4%** vs packed WARC |
|
| 175 |
|
| 176 |
+
The big win is the HTML → Markdown step: trafilatura strips all tags, scripts, styles, navigation, and ads, keeping only the main content. This cuts 1.9 TB of uncompressed HTML down to 70.8 GB of markdown — a **98.3% reduction** — before any file-level compression is applied. Parquet with Zstd level 19 then compresses the markdown a further 68.2%.
|
| 177 |
|
| 178 |
+
End to end: ~658.2 GB of raw gzipped WARCs becomes **22.5 GB of Parquet** — a **96.6% total reduction** — containing 15,708,074 clean markdown documents.
|
| 179 |
|
| 180 |
### Processing Times
|
| 181 |
|
| 182 |
+
Pipeline timings across 812 shards of CC-MAIN-2026-08:
|
| 183 |
|
| 184 |
```
|
| 185 |
Download (raw WARC) ████████████████████████ total 46h 34m 39s avg 3m 26s
|
| 186 |
+
Convert (HTML → MD) █░░░░░░░░░░░░░░░░░░░░░░░ total 3h 7m 11s avg 13s
|
| 187 |
+
Export (Parquet) █████░░░░░░░░░░░░░░░░░░░ total 10h 9m 20s avg 45s
|
| 188 |
Publish (HuggingFace) ███░░░░░░░░░░░░░░░░░░░░░ total 5h 59m 20s avg 26s
|
| 189 |
```
|
| 190 |
|
data/CC-MAIN-2026-08/00333.parquet
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:a63e6a89dae933b9325cd2e1a3c3cecdab2f22e53c72be548bba2919f7df19e7
|
| 3 |
+
size 29298182
|
stats.csv
CHANGED
|
@@ -255,6 +255,7 @@ CC-MAIN-2026-08,329,19490,2578092970,94257462,29579596,2026-03-13T15:22:02Z,283,
|
|
| 255 |
CC-MAIN-2026-08,330,19236,2590008930,92805723,30030776,2026-03-13T17:50:47Z,0,401,75,0
|
| 256 |
CC-MAIN-2026-08,331,19361,2559962643,92469515,29006296,2026-03-13T18:03:29Z,0,471,62,7
|
| 257 |
CC-MAIN-2026-08,332,19576,2569972796,95254532,30402651,2026-03-13T18:10:41Z,0,590,62,19
|
|
|
|
| 258 |
CC-MAIN-2026-08,401,19441,2586676962,94334130,29618260,2026-03-13T11:41:00Z,0,0,33,0
|
| 259 |
CC-MAIN-2026-08,402,19446,2552785517,96654791,30560442,2026-03-13T11:49:21Z,0,443,57,0
|
| 260 |
CC-MAIN-2026-08,403,19320,2579048144,91406292,29341198,2026-03-13T11:56:46Z,0,414,29,0
|
|
|
|
| 255 |
CC-MAIN-2026-08,330,19236,2590008930,92805723,30030776,2026-03-13T17:50:47Z,0,401,75,0
|
| 256 |
CC-MAIN-2026-08,331,19361,2559962643,92469515,29006296,2026-03-13T18:03:29Z,0,471,62,7
|
| 257 |
CC-MAIN-2026-08,332,19576,2569972796,95254532,30402651,2026-03-13T18:10:41Z,0,590,62,19
|
| 258 |
+
CC-MAIN-2026-08,333,19521,2571725088,93980385,29298182,2026-03-13T18:18:26Z,0,357,104,0
|
| 259 |
CC-MAIN-2026-08,401,19441,2586676962,94334130,29618260,2026-03-13T11:41:00Z,0,0,33,0
|
| 260 |
CC-MAIN-2026-08,402,19446,2552785517,96654791,30560442,2026-03-13T11:49:21Z,0,443,57,0
|
| 261 |
CC-MAIN-2026-08,403,19320,2579048144,91406292,29341198,2026-03-13T11:56:46Z,0,414,29,0
|