tamnd commited on
Commit
b046309
·
verified ·
1 Parent(s): 2dc5d12

Publish shard CC-MAIN-2026-08/00032

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 **640,760 documents across 32 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
 
@@ -166,27 +166,27 @@ Empty conversions (pages where trafilatura could not extract meaningful content)
166
 
167
  ### Compression Ratios
168
 
169
- Numbers below are actual measurements summed across all 32 files of CC-MAIN-2026-08 (640,760 pages total), projected to the full crawl of 100,000 WARC files.
170
 
171
- | Stage | 32 files (measured) | 100,000 files (projected) | Reduction |
172
  |---|---|---|---|
173
- | Raw WARC (.warc.gz, downloaded) | ~25.9 GB | ~83 TB | — |
174
- | HTML extracted (uncompressed) | 80.7 GB | ~295 TB | — |
175
- | Packed markdown WARC (.md.warc.gz) | ~1.2 GB | ~3.7 TB | **-98.5%** vs HTML |
176
- | Final Parquet (Zstd level 19) | 881.6 MB | ~2.9 TB | **-31.0%** vs packed WARC |
177
 
178
- The big win is the HTML → Markdown step: trafilatura strips all tags, scripts, styles, navigation, and ads, keeping only the main content. This cuts 80.7 GB of uncompressed HTML down to 2.7 GB of markdown — a **98.5% reduction** — before any file-level compression is applied. Parquet with Zstd level 19 then compresses the markdown a further 67.6%.
179
 
180
- End to end: ~25.9 GB of raw gzipped WARCs becomes **881.6 MB of Parquet** — a **96.7% total reduction** — containing 640,760 clean markdown documents.
181
 
182
  ### Processing Times
183
 
184
- Pipeline timings across 32 shards of CC-MAIN-2026-08:
185
 
186
  ```
187
- Pack (download + HTML→MD) ████████████████████████ total 7m 56s avg 14s
188
- Export (Parquet) █████░░░░░░░░░░░░░░░░░░░ total 1m 58s avg 3s
189
- Publish (HuggingFace) ██████░░░░░░░░░░░░░░░░░ total 2m 26s avg 4s
190
  ```
191
 
192
  ### Personal and Sensitive Information
 
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 **660,165 documents across 33 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
 
 
166
 
167
  ### Compression Ratios
168
 
169
+ Numbers below are actual measurements summed across all 33 files of CC-MAIN-2026-08 (660,165 pages total), projected to the full crawl of 100,000 WARC files.
170
 
171
+ | Stage | 33 files (measured) | 100,000 files (projected) | Reduction |
172
  |---|---|---|---|
173
+ | Raw WARC (.warc.gz, downloaded) | ~26.7 GB | ~83 TB | — |
174
+ | HTML extracted (uncompressed) | 83.0 GB | ~295 TB | — |
175
+ | Packed markdown WARC (.md.warc.gz) | ~1.3 GB | ~3.7 TB | **-98.4%** vs HTML |
176
+ | Final Parquet (Zstd level 19) | 910.5 MB | ~2.9 TB | **-31.0%** vs packed WARC |
177
 
178
+ The big win is the HTML → Markdown step: trafilatura strips all tags, scripts, styles, navigation, and ads, keeping only the main content. This cuts 83.0 GB of uncompressed HTML down to 2.7 GB of markdown — a **98.4% reduction** — before any file-level compression is applied. Parquet with Zstd level 19 then compresses the markdown a further 67.6%.
179
 
180
+ End to end: ~26.7 GB of raw gzipped WARCs becomes **910.5 MB of Parquet** — a **96.7% total reduction** — containing 660,165 clean markdown documents.
181
 
182
  ### Processing Times
183
 
184
+ Pipeline timings across 33 shards of CC-MAIN-2026-08:
185
 
186
  ```
187
+ Pack (download + HTML→MD) ████████████████████████ total 9m 25s avg 17s
188
+ Export (Parquet) █████░░░░░░░░░░░░░░░░░░░ total 2m 17s avg 4s
189
+ Publish (HuggingFace) ██████░░░░░░░░░░░░░░░░░ total 2m 28s avg 4s
190
  ```
191
 
192
  ### Personal and Sensitive Information
charts/compression_pie.png CHANGED
charts/rows_chart.png CHANGED
charts/size_chart.png CHANGED
charts/timing_chart.png CHANGED
data/CC-MAIN-2026-08/00032.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:acb7a36694aa2e35ad6fd173cfe61471e9eefad9c92055320d25a8bd2421467b
3
+ size 30315860
stats.csv CHANGED
@@ -30,4 +30,5 @@ CC-MAIN-2026-08,27,19612,2596439798,93518836,30208089,2026-03-13T05:52:18Z,71,18
30
  CC-MAIN-2026-08,28,19560,2579397903,97788121,30727702,2026-03-13T05:53:55Z,71,17,4
31
  CC-MAIN-2026-08,29,19439,2588405263,93704600,29839666,2026-03-13T05:55:32Z,72,17,2
32
  CC-MAIN-2026-08,30,19433,2598632809,94875082,30012372,2026-03-13T05:57:10Z,72,18,2
33
- CC-MAIN-2026-08,31,19468,2618291405,91912204,28962255,2026-03-13T05:58:42Z,75,16,0
 
 
30
  CC-MAIN-2026-08,28,19560,2579397903,97788121,30727702,2026-03-13T05:53:55Z,71,17,4
31
  CC-MAIN-2026-08,29,19439,2588405263,93704600,29839666,2026-03-13T05:55:32Z,72,17,2
32
  CC-MAIN-2026-08,30,19433,2598632809,94875082,30012372,2026-03-13T05:57:10Z,72,18,2
33
+ CC-MAIN-2026-08,31,19468,2618291405,91912204,28962255,2026-03-13T05:58:48Z,75,16,2
34
+ CC-MAIN-2026-08,32,19405,2562531576,93668802,30315860,2026-03-13T06:00:38Z,89,19,0