tamnd commited on
Commit
f038264
·
verified ·
1 Parent(s): 4bca02b

Publish shard CC-MAIN-2026-08/00922

Browse files
Files changed (3) hide show
  1. README.md +9 -9
  2. data/CC-MAIN-2026-08/00922.parquet +3 -0
  3. stats.csv +2 -1
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 **47,861,213 documents across 2470 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
 
@@ -164,28 +164,28 @@ Empty conversions (pages where trafilatura could not extract meaningful content)
164
 
165
  ### Compression Ratios
166
 
167
- Numbers below are actual measurements summed across all 2470 files of CC-MAIN-2026-08 (47,861,213 pages total), projected to the full crawl of 100,000 WARC files.
168
 
169
- | Stage | 2470 files (measured) | 100,000 files (projected) | Reduction |
170
  |---|---|---|---|
171
  | Raw WARC (.warc.gz, downloaded) | ~2.0 TB | ~83 TB | — |
172
  | HTML extracted (uncompressed) | 5.8 TB | ~295 TB | — |
173
  | Packed markdown WARC (.md.warc.gz) | ~101.2 GB | ~3.7 TB | **-98.3%** vs HTML |
174
  | Final Parquet (Zstd level 19) | 68.3 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 5.8 TB of uncompressed HTML down to 215.3 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: ~2.0 TB of raw gzipped WARCs becomes **68.3 GB of Parquet** — a **96.6% total reduction** — containing 47,861,213 clean markdown documents.
179
 
180
  ### Processing Times
181
 
182
- Pipeline timings across 2470 shards of CC-MAIN-2026-08:
183
 
184
  ```
185
- Download (raw WARC) ████████████████████████ total 213h 23m 0s avg 5m 11s
186
  Convert (HTML → MD) █░░░░░░░░░░░░░░░░░░░░░░░ total 12h 57m 19s avg 18s
187
- Export (Parquet) ██████░░░░░░░░░░░░░░░░░░ total 55h 0m 53s avg 1m 20s
188
- Publish (HuggingFace) █░░░░░░░░░░░░░░░░░░░░░░░ total 17h 37m 42s avg 25s
189
  ```
190
 
191
  ### Dataset Charts
 
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 **47,880,752 documents across 2471 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 2471 files of CC-MAIN-2026-08 (47,880,752 pages total), projected to the full crawl of 100,000 WARC files.
168
 
169
+ | Stage | 2471 files (measured) | 100,000 files (projected) | Reduction |
170
  |---|---|---|---|
171
  | Raw WARC (.warc.gz, downloaded) | ~2.0 TB | ~83 TB | — |
172
  | HTML extracted (uncompressed) | 5.8 TB | ~295 TB | — |
173
  | Packed markdown WARC (.md.warc.gz) | ~101.2 GB | ~3.7 TB | **-98.3%** vs HTML |
174
  | Final Parquet (Zstd level 19) | 68.3 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 5.8 TB of uncompressed HTML down to 215.4 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: ~2.0 TB of raw gzipped WARCs becomes **68.3 GB of Parquet** — a **96.6% total reduction** — containing 47,880,752 clean markdown documents.
179
 
180
  ### Processing Times
181
 
182
+ Pipeline timings across 2471 shards of CC-MAIN-2026-08:
183
 
184
  ```
185
+ Download (raw WARC) ████████████████████████ total 213h 35m 46s avg 5m 11s
186
  Convert (HTML → MD) █░░░░░░░░░░░░░░░░░░░░░░░ total 12h 57m 19s avg 18s
187
+ Export (Parquet) ██████░░░░░░░░░░░░░░░░░░ total 55h 7m 26s avg 1m 20s
188
+ Publish (HuggingFace) █░░░░░░░░░░░░░░░░░░░░░░░ total 17h 38m 0s avg 25s
189
  ```
190
 
191
  ### Dataset Charts
data/CC-MAIN-2026-08/00922.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e7916c455e969539acbbc1ffd56cabb15f564158486dd824ff72d6d197c85374
3
+ size 29762394
stats.csv CHANGED
@@ -644,7 +644,7 @@ CC-MAIN-2026-08,657,19219,2538250599,94251414,29215461,2026-03-14T15:31:16Z,1048
644
  CC-MAIN-2026-08,658,19294,2556889298,95437715,30831548,2026-03-14T15:47:52Z,868,0,121,25
645
  CC-MAIN-2026-08,659,19558,2616128595,94371966,29524419,2026-03-14T16:12:37Z,1216,0,231,16
646
  CC-MAIN-2026-08,660,19380,2540748926,97827507,30882600,2026-03-14T16:31:38Z,782,0,400,19
647
- CC-MAIN-2026-08,661,19372,2563357995,93912950,29524634,2026-03-14T16:50:38Z,668,0,399,0
648
  CC-MAIN-2026-08,701,19575,2592915946,93927019,29556184,2026-03-13T11:40:57Z,0,0,30,0
649
  CC-MAIN-2026-08,702,19449,2572639558,92293569,29617754,2026-03-13T11:49:10Z,0,432,59,0
650
  CC-MAIN-2026-08,703,19592,2602952509,95185029,30715031,2026-03-13T11:56:07Z,0,318,97,0
@@ -858,6 +858,7 @@ CC-MAIN-2026-08,918,19597,2580097026,94188557,29611897,2026-03-14T16:12:37Z,1208
858
  CC-MAIN-2026-08,919,19253,2556548461,95660440,29984976,2026-03-14T16:32:45Z,814,0,428,19
859
  CC-MAIN-2026-08,920,19369,2571940451,97880180,31397573,2026-03-13T13:53:05Z,0,174,60,0
860
  CC-MAIN-2026-08,921,19336,2562062820,94054773,30754989,2026-03-13T13:59:15Z,338,0,29,0
 
861
  CC-MAIN-2026-08,930,19425,2563571878,93507547,29730912,2026-03-13T14:07:36Z,198,0,21,0
862
  CC-MAIN-2026-08,1000,19490,2581039608,93029785,29406839,2026-03-13T11:28:30Z,251,0,36,0
863
  CC-MAIN-2026-08,1001,19530,2577623205,97623904,31829877,2026-03-13T10:02:20Z,605,0,422,0
 
644
  CC-MAIN-2026-08,658,19294,2556889298,95437715,30831548,2026-03-14T15:47:52Z,868,0,121,25
645
  CC-MAIN-2026-08,659,19558,2616128595,94371966,29524419,2026-03-14T16:12:37Z,1216,0,231,16
646
  CC-MAIN-2026-08,660,19380,2540748926,97827507,30882600,2026-03-14T16:31:38Z,782,0,400,19
647
+ CC-MAIN-2026-08,661,19372,2563357995,93912950,29524634,2026-03-14T16:50:38Z,668,0,399,18
648
  CC-MAIN-2026-08,701,19575,2592915946,93927019,29556184,2026-03-13T11:40:57Z,0,0,30,0
649
  CC-MAIN-2026-08,702,19449,2572639558,92293569,29617754,2026-03-13T11:49:10Z,0,432,59,0
650
  CC-MAIN-2026-08,703,19592,2602952509,95185029,30715031,2026-03-13T11:56:07Z,0,318,97,0
 
858
  CC-MAIN-2026-08,919,19253,2556548461,95660440,29984976,2026-03-14T16:32:45Z,814,0,428,19
859
  CC-MAIN-2026-08,920,19369,2571940451,97880180,31397573,2026-03-13T13:53:05Z,0,174,60,0
860
  CC-MAIN-2026-08,921,19336,2562062820,94054773,30754989,2026-03-13T13:59:15Z,338,0,29,0
861
+ CC-MAIN-2026-08,922,19539,2532904600,91826410,29762394,2026-03-14T16:52:07Z,766,0,393,0
862
  CC-MAIN-2026-08,930,19425,2563571878,93507547,29730912,2026-03-13T14:07:36Z,198,0,21,0
863
  CC-MAIN-2026-08,1000,19490,2581039608,93029785,29406839,2026-03-13T11:28:30Z,251,0,36,0
864
  CC-MAIN-2026-08,1001,19530,2577623205,97623904,31829877,2026-03-13T10:02:20Z,605,0,422,0