Update README.md
Browse files
README.md
CHANGED
|
@@ -21,26 +21,34 @@ Corpus-200B/
|
|
| 21 |
documents/ # Pre-processed web documents
|
| 22 |
- CC_shard_00000000_processed.jsonl.zst
|
| 23 |
- CC_shard_00000001_processed.jsonl.zst
|
|
|
|
| 24 |
- ...
|
| 25 |
tokens/ # number of tokens per document (GPT-NeoX tokenizer)
|
| 26 |
- CC_shard_00000000_processed.npy
|
| 27 |
- CC_shard_00000001_processed.npy
|
|
|
|
| 28 |
- ...
|
| 29 |
scores_dclm-fasttext/ # DCLM-fasttext score
|
| 30 |
- CC_shard_00000000_processed.npy
|
| 31 |
- ...
|
| 32 |
scores_fineweb-edu/ # FineWeb-Edu score
|
| 33 |
- CC_shard_00000000_processed.npy
|
| 34 |
-
-
|
|
|
|
|
|
|
| 35 |
- ...
|
| 36 |
domains_topics/ # TopicClassifier annotations
|
| 37 |
-
- CC_shard_00000000_processed__logits.npy # logits for each topic
|
| 38 |
- CC_shard_00000000_processed__choice.npy # index of top choice
|
| 39 |
- ...
|
|
|
|
|
|
|
|
|
|
| 40 |
domains_formats/ # FormatClassifier annotations
|
| 41 |
-
- CC_shard_00000000_processed__logits.npy # logits for each format
|
| 42 |
- CC_shard_00000000_processed__choice.npy # index of top choice
|
| 43 |
- ...
|
|
|
|
|
|
|
|
|
|
| 44 |
domains_clusters-k24/ # K-means clusters
|
| 45 |
- CC_shard_00000000_processed.npy # cluster assignment for each document
|
| 46 |
- ...
|
|
@@ -54,6 +62,7 @@ If you make use of this pre-processed corpus in your work, please cite:
|
|
| 54 |
@article{wettig2025organize,
|
| 55 |
title={Organize the Web: Constructing Domains Enhances Pre-Training Data Curation},
|
| 56 |
author={Alexander Wettig and Kyle Lo and Sewon Min and Hannaneh Hajishirzi and Danqi Chen and Luca Soldaini},
|
|
|
|
| 57 |
year={2025}
|
| 58 |
}
|
| 59 |
```
|
|
|
|
| 21 |
documents/ # Pre-processed web documents
|
| 22 |
- CC_shard_00000000_processed.jsonl.zst
|
| 23 |
- CC_shard_00000001_processed.jsonl.zst
|
| 24 |
+
- CC_shard_00000002_processed.jsonl.zst
|
| 25 |
- ...
|
| 26 |
tokens/ # number of tokens per document (GPT-NeoX tokenizer)
|
| 27 |
- CC_shard_00000000_processed.npy
|
| 28 |
- CC_shard_00000001_processed.npy
|
| 29 |
+
- CC_shard_00000002_processed.npy
|
| 30 |
- ...
|
| 31 |
scores_dclm-fasttext/ # DCLM-fasttext score
|
| 32 |
- CC_shard_00000000_processed.npy
|
| 33 |
- ...
|
| 34 |
scores_fineweb-edu/ # FineWeb-Edu score
|
| 35 |
- CC_shard_00000000_processed.npy
|
| 36 |
+
- ...
|
| 37 |
+
scores_fineweb-edu__rounded/ # Rounded FineWeb-Edu score
|
| 38 |
+
- CC_shard_00000000_processed__rounded.npy
|
| 39 |
- ...
|
| 40 |
domains_topics/ # TopicClassifier annotations
|
|
|
|
| 41 |
- CC_shard_00000000_processed__choice.npy # index of top choice
|
| 42 |
- ...
|
| 43 |
+
domain_topics__logits/
|
| 44 |
+
- CC_shard_00000000_processed__logits.npy # logits for each topic
|
| 45 |
+
- ...
|
| 46 |
domains_formats/ # FormatClassifier annotations
|
|
|
|
| 47 |
- CC_shard_00000000_processed__choice.npy # index of top choice
|
| 48 |
- ...
|
| 49 |
+
domains_formats/ # FormatClassifier annotations
|
| 50 |
+
- CC_shard_00000000_processed__logits.npy # logits for each format
|
| 51 |
+
- ...
|
| 52 |
domains_clusters-k24/ # K-means clusters
|
| 53 |
- CC_shard_00000000_processed.npy # cluster assignment for each document
|
| 54 |
- ...
|
|
|
|
| 62 |
@article{wettig2025organize,
|
| 63 |
title={Organize the Web: Constructing Domains Enhances Pre-Training Data Curation},
|
| 64 |
author={Alexander Wettig and Kyle Lo and Sewon Min and Hannaneh Hajishirzi and Danqi Chen and Luca Soldaini},
|
| 65 |
+
journal={arXiv preprint arXiv:2502.10341},
|
| 66 |
year={2025}
|
| 67 |
}
|
| 68 |
```
|