tamnd commited on
Commit
c926a4e
·
verified ·
1 Parent(s): 13547c1

Publish 2 shards CC-MAIN-2026-08/00593–01335

Browse files
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 **25,228,420 documents across 1305 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
 
@@ -164,27 +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 1305 files of CC-MAIN-2026-08 (25,228,420 pages total), projected to the full crawl of 100,000 WARC files.
168
 
169
- | Stage | 1305 files (measured) | 100,000 files (projected) | Reduction |
170
  |---|---|---|---|
171
  | Raw WARC (.warc.gz, downloaded) | ~1.0 TB | ~83 TB | — |
172
  | HTML extracted (uncompressed) | 3.0 TB | ~295 TB | — |
173
- | Packed markdown WARC (.md.warc.gz) | ~53.5 GB | ~3.7 TB | **-98.3%** vs HTML |
174
- | Final Parquet (Zstd level 19) | 36.1 GB | ~2.9 TB | **-32.5%** 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 3.0 TB of uncompressed HTML down to 113.8 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 68.3%.
177
 
178
- End to end: ~1.0 TB of raw gzipped WARCs becomes **36.1 GB of Parquet** — a **96.6% total reduction** — containing 25,228,420 clean markdown documents.
179
 
180
  ### Processing Times
181
 
182
- Pipeline timings across 1305 shards of CC-MAIN-2026-08:
183
 
184
  ```
185
- Download (raw WARC) ████████████████████████ total 100h 38m 20s avg 4m 37s
186
- Convert (HTML → MD) █░░░░░░░░░░░░░░░░░░░░░░░ total 6h 51m 14s avg 18s
187
- Export (Parquet) █████░░░░░░░░░░░░░░░░░░░ total 24h 29m 38s avg 1m 7s
188
  Publish (HuggingFace) █░░░░░░░░░░░░░░░░░░░░░░░ total 8h 7m 44s avg 22s
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 **25,267,164 documents across 1307 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 1307 files of CC-MAIN-2026-08 (25,267,164 pages total), projected to the full crawl of 100,000 WARC files.
168
 
169
+ | Stage | 1307 files (measured) | 100,000 files (projected) | Reduction |
170
  |---|---|---|---|
171
  | Raw WARC (.warc.gz, downloaded) | ~1.0 TB | ~83 TB | — |
172
  | HTML extracted (uncompressed) | 3.0 TB | ~295 TB | — |
173
+ | Packed markdown WARC (.md.warc.gz) | ~53.6 GB | ~3.7 TB | **-98.3%** vs HTML |
174
+ | Final Parquet (Zstd level 19) | 36.2 GB | ~2.9 TB | **-32.5%** 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 3.0 TB of uncompressed HTML down to 114.0 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 68.3%.
177
 
178
+ End to end: ~1.0 TB of raw gzipped WARCs becomes **36.2 GB of Parquet** — a **96.6% total reduction** — containing 25,267,164 clean markdown documents.
179
 
180
  ### Processing Times
181
 
182
+ Pipeline timings across 1307 shards of CC-MAIN-2026-08:
183
 
184
  ```
185
+ Download (raw WARC) ████████████████████████ total 100h 46m 23s avg 4m 37s
186
+ Convert (HTML → MD) █░░░░░░░░░░░░░░░░░░░░░░░ total 6h 59m 32s avg 19s
187
+ Export (Parquet) █████░░░░░░░░░░░░░░░░░░░ total 24h 33m 52s avg 1m 7s
188
  Publish (HuggingFace) █░░░░░░░░░░░░░░░░░░░░░░░ total 8h 7m 44s avg 22s
189
  ```
190
 
data/CC-MAIN-2026-08/00593.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:16afedd5d35d2d97056966e9e205c98c192e509db57821714946ef009ad6a8ec
3
+ size 30115196
data/CC-MAIN-2026-08/01335.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f935ded1883b0c1181fcccb02f66f241aba52b36b721064112c9f720016d89d0
3
+ size 29547506
stats.csv CHANGED
@@ -494,6 +494,7 @@ CC-MAIN-2026-08,589,19258,2537999141,94874223,30339310,2026-03-14T02:38:05Z,434,
494
  CC-MAIN-2026-08,590,19483,2570363266,98986063,31831394,2026-03-14T02:49:50Z,0,545,161,15
495
  CC-MAIN-2026-08,591,19537,2561813555,95325642,29831248,2026-03-14T02:59:36Z,0,345,219,29
496
  CC-MAIN-2026-08,592,19444,2584070335,97424419,30482871,2026-03-14T03:08:41Z,0,475,88,38
 
497
  CC-MAIN-2026-08,600,19705,2589549495,96308923,30604838,2026-03-13T13:52:57Z,0,170,56,43
498
  CC-MAIN-2026-08,601,19595,2603790636,92061417,29528881,2026-03-13T13:58:00Z,178,0,123,43
499
  CC-MAIN-2026-08,602,19383,2595459366,97340911,31273659,2026-03-13T14:00:42Z,114,0,47,43
@@ -889,6 +890,7 @@ CC-MAIN-2026-08,1331,19502,2578969498,98418891,31188964,2026-03-14T02:35:20Z,466
889
  CC-MAIN-2026-08,1332,19240,2578734476,96883680,30443777,2026-03-14T02:47:02Z,569,0,132,14
890
  CC-MAIN-2026-08,1333,19398,2611873016,95906282,30349927,2026-03-14T02:58:59Z,477,0,61,31
891
  CC-MAIN-2026-08,1334,19616,2625127244,98361569,30410115,2026-03-14T03:10:17Z,666,0,95,14
 
892
  CC-MAIN-2026-08,1401,19441,2586676962,94334130,29606093,2026-03-13T09:50:34Z,230,0,90,0
893
  CC-MAIN-2026-08,1402,19641,2571912489,94392153,29376558,2026-03-13T10:21:08Z,1035,0,797,0
894
  CC-MAIN-2026-08,1403,19445,2589553842,99835858,31697837,2026-03-13T10:51:00Z,1248,0,530,0
 
494
  CC-MAIN-2026-08,590,19483,2570363266,98986063,31831394,2026-03-14T02:49:50Z,0,545,161,15
495
  CC-MAIN-2026-08,591,19537,2561813555,95325642,29831248,2026-03-14T02:59:36Z,0,345,219,29
496
  CC-MAIN-2026-08,592,19444,2584070335,97424419,30482871,2026-03-14T03:08:41Z,0,475,88,38
497
+ CC-MAIN-2026-08,593,19299,2567619417,94185300,30115196,2026-03-14T03:19:19Z,0,498,138,0
498
  CC-MAIN-2026-08,600,19705,2589549495,96308923,30604838,2026-03-13T13:52:57Z,0,170,56,43
499
  CC-MAIN-2026-08,601,19595,2603790636,92061417,29528881,2026-03-13T13:58:00Z,178,0,123,43
500
  CC-MAIN-2026-08,602,19383,2595459366,97340911,31273659,2026-03-13T14:00:42Z,114,0,47,43
 
890
  CC-MAIN-2026-08,1332,19240,2578734476,96883680,30443777,2026-03-14T02:47:02Z,569,0,132,14
891
  CC-MAIN-2026-08,1333,19398,2611873016,95906282,30349927,2026-03-14T02:58:59Z,477,0,61,31
892
  CC-MAIN-2026-08,1334,19616,2625127244,98361569,30410115,2026-03-14T03:10:17Z,666,0,95,14
893
+ CC-MAIN-2026-08,1335,19445,2547826342,94548063,29547506,2026-03-14T03:19:19Z,483,0,116,0
894
  CC-MAIN-2026-08,1401,19441,2586676962,94334130,29606093,2026-03-13T09:50:34Z,230,0,90,0
895
  CC-MAIN-2026-08,1402,19641,2571912489,94392153,29376558,2026-03-13T10:21:08Z,1035,0,797,0
896
  CC-MAIN-2026-08,1403,19445,2589553842,99835858,31697837,2026-03-13T10:51:00Z,1248,0,530,0