Hugging Face's logo

Datasets:
togethercomputer
/
RedPajama-Data-V2

Tasks:
Text Generation
Modalities:
Text
Languages:
English
German
French
Size:
1M<n<10M
ArXiv:
Tags:
Libraries:
Datasets
Croissant
Dataset card Viewer Files Community
28
Maurice Weber commited on
Commit
f5f282b
•
1 Parent(s): e43cbf3

update README.md

Browse files
Files changed (1) hide show
  1. README.md +87 -56
README.md CHANGED
@@ -7,7 +7,7 @@ language:
7
  - fr
8
  - es
9
  - it
10
- pretty_name: Red Pajama V2 Data Foundation
11
  ---
12
 
13
  ### Getting Started
@@ -83,56 +83,61 @@ RedPajama-V2 is an open dataset for training large laguage models and includes o
83
 
84
  #### Quality Annotations
85
 
86
- | Annotation Tag | Description | Category | Reference |
87
- |------------------------------------------------|----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|------------------|------------------------------------|
88
- | ccnet_bucket | head, middle or tail bucket of the perplexity score | ccnet | ccnet |
89
- | ccnet_language_score | score of the language identification model | ccnet | ccnet |
90
- | ccnet_length | number of characters | ccnet | ccnet |
91
- | ccnet_nlines | number of lines | ccnet | ccnet |
92
- | ccnet_original_length | number of characters before in-document line deduplication | ccnet | ccnet |
93
- | ccnet_original_nlines | number of lines before in-document line deduplication | ccnet | ccnet |
94
- | ccnet_perplexity | perplexity of an LM trained on Wikipedia | ccnet | ccnet |
95
- | rps_doc_books_importance | Given a bag of {1,2}-wordgram model trained on Books p, and a model trained on the source domain q, This is the logarithm of the ratio p(doc)/q(doc). | ML Heuristics | Importance Resampling (Xie et al.) |
96
- | rps_doc_openwebtext_importance | Given a bag of {1,2}-wordgram model trained on OpenWebText p, and a model trained on the source domain q, this is the logarithm of the ratio p(doc)/q(doc). | ML Heuristics | Importance Resampling (Xie et al.) |
97
- | rps_doc_wikipedia_importance | Given a bag of {1,2}-wordgram model trained on Wikipedia articles p, and a model trained on the source domain q, this is the logarithm of the ratio p(doc)/q(doc). | ML Heuristics | Importance Resampling (Xie et al.) |
98
- | rps_doc_ml_wikiref_score | Fasttext classifier prediction for the document being a Wikipedia reference. This is the same fasttext model used in the RedPajama-1T dataset. Only applies to English data.. | ML Heuristics | LLaMA, RedPajama-1T |
99
- | rps_doc_ml_palm_score | Fasttext classifier prediction for the document being a Wikipedia article, OpenWebText sample or a RedPajama-V1 book. Only for English data. | ML Heuristics | PaLM, GLaM |
100
- | rps_doc_ml_wikipedia_score | Fasttext classifier prediction for the document being a Wikipedia article. This is used for non-English data | ML Heuristics | - |
101
- | rps_doc_curly_bracket | The ratio between the number of occurrences of '{' or '}' and the number of characters in the raw text. | Natural Language | C4 |
102
- | rps_doc_frac_all_caps_words | The fraction of words in the content that only consist of uppercase letters. This is based on the raw content. | Natural Language | Pretrainer’s Guide |
103
- | rps_doc_frac_lines_end_with_ellipsis | The fraction of lines that end with an ellipsis, where an ellipsis is defined as either "..." or "…". | Natural Language | RefinedWeb, Gopher |
104
- | rps_doc_frac_no_alph_words | The fraction of words that contain no alphabetical character. | Natural Language | RefinedWeb, Gopher |
105
- | rps_doc_lorem_ipsum | The ratio between the number of occurrences of 'lorem ipsum' and the number of characters in the content after normalisation. | Natural Language | C4 |
106
- | rps_doc_mean_word_length | The mean length of words in the content after normalisation. | Natural Language | RefinedWeb, Gopher |
107
- | rps_doc_stop_word_fraction | The ratio between the number of stop words and the number of words in the document. Stop words are obtained from https://github.com/6/stopwords-json. | Natural Language | RefinedWeb, Gopher |
108
- | rps_doc_symbol_to_word_ratio | The ratio of symbols to words in the content.. Symbols are defined "#", "...", and "…". | Natural Language | RefinedWeb, Gopher |
109
- | rps_doc_frac_unique_words | The fraction of unique words in the content. This is also known as the degeneracy of a text sample. Calculated based on the normalised content. | Natural Language | Pretrainer’s Guide |
110
- | rps_doc_unigram_entropy | The entropy of the unigram distribution of the content. This measures the diversity of the content and is computed using sum(-x / total * log(x / total)) where the sum is taken over counts of unique words in the normalised content. | Natural Language | - |
111
- | rps_doc_word_count | The number of words in the content after normalisation. | Natural Language | RefinedWeb, Gopher |
112
- | rps_lines_ending_with_terminal_punctution_mark | Indicates whether a line ends with a terminal punctuation mark. A terminal punctation mark is defined as one of: ".", "!", "?", "”". | Natural Language | C4 |
113
- | rps_lines_javascript_counts | The number of occurrences of the word "javascript" in each line. | Natural Language | C4 |
114
- | rps_lines_num_words | The number of words in each line. This is computed based on the normalised text. | Natural Language | C4, RefinedWeb |
115
- | rps_lines_numerical_chars_fraction | The ratio between the number of numerical characters and total number of characters in each line. This is based on the normalised content. | Natural Language | RefinedWeb |
116
- | rps_lines_start_with_bulletpoint | Whether the lines that start with a bullet point symbol. The following set of unicodes are considered a bullet point: \u2022 (bullet point), \u2023 (triangular bullet point), \u25B6 (black right pointing triangle), \u25C0 (black left pointing triangle), \u25E6 (white bullet point), \u25A0 (black square), \u25A1 (white square), \u25AA (black small square), \u25AB (white small square), \u2013 (en dash). | Natural Language | RefinedWeb, Gopher |
117
- | rps_lines_uppercase_letter_fraction | The ratio between the number of uppercase letters and total number of characters in each line. This is based on the raw text. | Natural Language | RefinedWeb |
118
- | rps_doc_num_sentences | The number of sentences in the content. This is calculated using the regular expression `r'\b[^.!?]+[.!?]*'`. | Natural Language | C4 |
119
- | rps_doc_frac_chars_dupe_10grams | The fraction of characters in duplicate word 10grams. This operates on the lower-cased, punctuation removed content. It is also ensured that characters in overlapping ngrams are only counted once. | Repetitiveness | RefinedWeb, Gopher |
120
- | rps_doc_frac_chars_dupe_5grams | The fraction of characters in duplicate word 5grams. | Repetitiveness | RefinedWeb, Gopher |
121
- | rps_doc_frac_chars_dupe_6grams | The fraction of characters in duplicate word 6grams. | Repetitiveness | RefinedWeb, Gopher |
122
- | rps_doc_frac_chars_dupe_7grams | The fraction of characters in duplicate word 7grams. | Repetitiveness | RefinedWeb, Gopher |
123
- | rps_doc_frac_chars_dupe_8grams | The fraction of characters in duplicate word 8grams. | Repetitiveness | RefinedWeb, Gopher |
124
- | rps_doc_frac_chars_dupe_9grams | The fraction of characters in duplicate word 9grams. | Repetitiveness | RefinedWeb, Gopher |
125
- | rps_doc_frac_chars_top_2gram | The fraction of characters in the top word 2gram. | Repetitiveness | RefinedWeb, Gopher |
126
- | rps_doc_frac_chars_top_3gram | The fraction of characters in the top word 3gram. | Repetitiveness | RefinedWeb, Gopher |
127
- | rps_doc_frac_chars_top_4gram | The fraction of characters in the top word 4gram. | Repetitiveness | RefinedWeb, Gopher |
128
- | rps_doc_ldnoobw_words | The number of sequences of words that are contained in the List-of-Dirty-Naughty-Obscene-and-Otherwise-Bad-Words blocklist. The blocklist is obtained from [GitHub](https://github.com/LDNOOBW/List-of-Dirty-Naughty-Obscene-and-Otherwise-Bad-Words). | toxicity | C4 |
129
- | rps_doc_ut1_blacklist | A categorical id corresponding to the list of categories of the domain of the document. Categories are obtained from the UT1 blacklist. The list is obtained from [UT-Capitole](https://dsi.ut-capitole.fr/blacklists/). | toxicictiy | RefinedWeb |
130
-
131
- #### Document Counts for the Annotated part of the dataset
132
-
133
- | | en | de | fr | es | it | Total |
134
- |-------------|-------|------|------|------|------|-------|
135
- | # Documents | 24.5B | 2.7B | 2.2B | 2.3B | 1.2B | 32.9B |
 
 
 
 
 
136
 
137
  ### Languages
138
 
@@ -169,7 +174,30 @@ The dataset is structured into four components, each following the same key stru
169
  ├── it_middle.minhash.parquet
170
  ```
171
 
172
- Documents files, which contain the text, folow the schema defined by CCNet, and the quality signals follow the schema
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
173
 
174
  ```json
175
  {
@@ -217,12 +245,15 @@ Documents files, which contain the text, folow the schema defined by CCNet, and
217
  ```
218
 
219
  where signal scores are encoded as a list of tuples `(start, end, score)`, where `start` and `end` are the locations in
220
- the
221
- `raw_content` string where the `score` applies.
222
 
223
  ## Dataset Creation
224
 
225
- The dataset is based on 84 snapshots provided by Common Crawl.
 
 
 
 
226
 
227
  ## Citation
228
 
@@ -231,7 +262,7 @@ To cite RedPajama-V2, please use:
231
  ```
232
  @software{together2023redpajama-v2,
233
  author = {Together Computer},
234
- title = {RedPajama-Data-v2: a living data foundation for training open LLM models},
235
  month = October,
236
  year = 2023,
237
  url = {https://github.com/togethercomputer/RedPajama-Data}
 
7
  - fr
8
  - es
9
  - it
10
+ pretty_name: Red Pajama V2 Dataset
11
  ---
12
 
13
  ### Getting Started
 
83
 
84
  #### Quality Annotations
85
 
86
+ | Annotation Tag | Description | Category | Reference |
87
+ |------------------------------------------------|----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|------------------|-------------------------------------------------------------------------------------------------------------------------------|
88
+ | ccnet_bucket | head, middle or tail bucket of the perplexity score | CCNet | [CCNet](https://github.com/facebookresearch/cc_net) |
89
+ | ccnet_language_score | score of the language identification model | CCNet | [CCNet](https://github.com/facebookresearch/cc_net) |
90
+ | ccnet_length | number of characters | CCNet | [CCNet](https://github.com/facebookresearch/cc_net) |
91
+ | ccnet_nlines | number of lines | CCNet | [CCNet](https://github.com/facebookresearch/cc_net) |
92
+ | ccnet_original_length | number of characters before in-document line deduplication | CCNet | [CCNet](https://github.com/facebookresearch/cc_net) |
93
+ | ccnet_original_nlines | number of lines before in-document line deduplication | CCNet | [CCNet](https://github.com/facebookresearch/cc_net) |
94
+ | ccnet_perplexity | perplexity of an LM trained on Wikipedia | CCNet | [CCNet](https://github.com/facebookresearch/cc_net) |
95
+ | rps_doc_books_importance | Given a bag of {1,2}-wordgram model trained on Books p, and a model trained on the source domain q, This is the logarithm of the ratio p(doc)/q(doc). | ML Heuristics | [Importance Resampling (Xie et al.)](https://arxiv.org/abs/2302.03169) |
96
+ | rps_doc_openwebtext_importance | Given a bag of {1,2}-wordgram model trained on OpenWebText p, and a model trained on the source domain q, this is the logarithm of the ratio p(doc)/q(doc). | ML Heuristics | [Importance Resampling (Xie et al.)](https://arxiv.org/abs/2302.03169) |
97
+ | rps_doc_wikipedia_importance | Given a bag of {1,2}-wordgram model trained on Wikipedia articles p, and a model trained on the source domain q, this is the logarithm of the ratio p(doc)/q(doc). | ML Heuristics | [Importance Resampling (Xie et al.)](https://arxiv.org/abs/2302.03169) |
98
+ | rps_doc_ml_wikiref_score | Fasttext classifier prediction for the document being a Wikipedia reference. This is the same fasttext model used in the RedPajama-1T dataset. Only applies to English data.. | ML Heuristics | [LLaMA](https://arxiv.org/abs/2302.13971), [RedPajama-1T](https://huggingface.co/datasets/togethercomputer/RedPajama-Data-1T) |
99
+ | rps_doc_ml_palm_score | Fasttext classifier prediction for the document being a Wikipedia article, OpenWebText sample or a RedPajama-V1 book. Only for English data. | ML Heuristics | [PALM](https://arxiv.org/abs/2204.02311), [GLaM](https://arxiv.org/abs/2112.06905) |
100
+ | rps_doc_ml_wikipedia_score | Fasttext classifier prediction for the document being a Wikipedia article. This is used for non-English data | ML Heuristics | - |
101
+ | rps_doc_curly_bracket | The ratio between the number of occurrences of '{' or '}' and the number of characters in the raw text. | Natural Language | [C4](https://arxiv.org/abs/1910.10683) |
102
+ | rps_doc_frac_all_caps_words | The fraction of words in the content that only consist of uppercase letters. This is based on the raw content. | Natural Language | [Pretrainer’s Guide](https://arxiv.org/abs/2305.13169) |
103
+ | rps_doc_frac_lines_end_with_ellipsis | The fraction of lines that end with an ellipsis, where an ellipsis is defined as either "..." or "…". | Natural Language | [RefinedWeb](https://arxiv.org/abs/2306.01116), [Gopher](https://arxiv.org/abs/2112.11446) |
104
+ | rps_doc_frac_no_alph_words | The fraction of words that contain no alphabetical character. | Natural Language | [RefinedWeb](https://arxiv.org/abs/2306.01116), [Gopher](https://arxiv.org/abs/2112.11446) |
105
+ | rps_doc_lorem_ipsum | The ratio between the number of occurrences of 'lorem ipsum' and the number of characters in the content after normalisation. | Natural Language | [C4](https://arxiv.org/abs/1910.10683) |
106
+ | rps_doc_mean_word_length | The mean length of words in the content after normalisation. | Natural Language | [RefinedWeb](https://arxiv.org/abs/2306.01116), [Gopher](https://arxiv.org/abs/2112.11446) |
107
+ | rps_doc_stop_word_fraction | The ratio between the number of stop words and the number of words in the document. Stop words are obtained from the [stopwords-json](https://github.com/6/stopwords-json) repo. | Natural Language | [RefinedWeb](https://arxiv.org/abs/2306.01116), [Gopher](https://arxiv.org/abs/2112.11446) |
108
+ | rps_doc_symbol_to_word_ratio | The ratio of symbols to words in the content.. Symbols are defined "#", "...", and "…". | Natural Language | [RefinedWeb](https://arxiv.org/abs/2306.01116), [Gopher](https://arxiv.org/abs/2112.11446) |
109
+ | rps_doc_frac_unique_words | The fraction of unique words in the content. This is also known as the degeneracy of a text sample. Calculated based on the normalised content. | Natural Language | [Pretrainer’s Guide](https://arxiv.org/abs/2305.13169) |
110
+ | rps_doc_unigram_entropy | The entropy of the unigram distribution of the content. This measures the diversity of the content and is computed using sum(-x / total * log(x / total)) where the sum is taken over counts of unique words in the normalised content. | Natural Language | - |
111
+ | rps_doc_word_count | The number of words in the content after normalisation. | Natural Language | [RefinedWeb](https://arxiv.org/abs/2306.01116), [Gopher](https://arxiv.org/abs/2112.11446) |
112
+ | rps_lines_ending_with_terminal_punctution_mark | Indicates whether a line ends with a terminal punctuation mark. A terminal punctation mark is defined as one of: ".", "!", "?", "”". | Natural Language | [C4](https://arxiv.org/abs/1910.10683) |
113
+ | rps_lines_javascript_counts | The number of occurrences of the word "javascript" in each line. | Natural Language | [C4](https://arxiv.org/abs/1910.10683) |
114
+ | rps_lines_num_words | The number of words in each line. This is computed based on the normalised text. | Natural Language | [C4](https://arxiv.org/abs/1910.10683) , [RefinedWeb](https://arxiv.org/abs/2306.01116) |
115
+ | rps_lines_numerical_chars_fraction | The ratio between the number of numerical characters and total number of characters in each line. This is based on the normalised content. | Natural Language | [RefinedWeb](https://arxiv.org/abs/2306.01116) |
116
+ | rps_lines_start_with_bulletpoint | Whether the lines that start with a bullet point symbol. The following set of unicodes are considered a bullet point: \u2022 (bullet point), \u2023 (triangular bullet point), \u25B6 (black right pointing triangle), \u25C0 (black left pointing triangle), \u25E6 (white bullet point), \u25A0 (black square), \u25A1 (white square), \u25AA (black small square), \u25AB (white small square), \u2013 (en dash). | Natural Language | [RefinedWeb](https://arxiv.org/abs/2306.01116), [Gopher](https://arxiv.org/abs/2112.11446) |
117
+ | rps_lines_uppercase_letter_fraction | The ratio between the number of uppercase letters and total number of characters in each line. This is based on the raw text. | Natural Language | [RefinedWeb](https://arxiv.org/abs/2306.01116) |
118
+ | rps_doc_num_sentences | The number of sentences in the content. This is calculated using the regular expression `r'\b[^.!?]+[.!?]*'`. | Natural Language | [C4](https://arxiv.org/abs/1910.10683) |
119
+ | rps_doc_frac_chars_dupe_10grams | The fraction of characters in duplicate word 10grams. This operates on the lower-cased, punctuation removed content. It is also ensured that characters in overlapping ngrams are only counted once. | Repetitiveness | [RefinedWeb](https://arxiv.org/abs/2306.01116), [Gopher](https://arxiv.org/abs/2112.11446) |
120
+ | rps_doc_frac_chars_dupe_5grams | The fraction of characters in duplicate word 5grams. | Repetitiveness | [RefinedWeb](https://arxiv.org/abs/2306.01116), [Gopher](https://arxiv.org/abs/2112.11446) |
121
+ | rps_doc_frac_chars_dupe_6grams | The fraction of characters in duplicate word 6grams. | Repetitiveness | [RefinedWeb](https://arxiv.org/abs/2306.01116), [Gopher](https://arxiv.org/abs/2112.11446) |
122
+ | rps_doc_frac_chars_dupe_7grams | The fraction of characters in duplicate word 7grams. | Repetitiveness | [RefinedWeb](https://arxiv.org/abs/2306.01116), [Gopher](https://arxiv.org/abs/2112.11446) |
123
+ | rps_doc_frac_chars_dupe_8grams | The fraction of characters in duplicate word 8grams. | Repetitiveness | [RefinedWeb](https://arxiv.org/abs/2306.01116), [Gopher](https://arxiv.org/abs/2112.11446) |
124
+ | rps_doc_frac_chars_dupe_9grams | The fraction of characters in duplicate word 9grams. | Repetitiveness | [RefinedWeb](https://arxiv.org/abs/2306.01116), [Gopher](https://arxiv.org/abs/2112.11446) |
125
+ | rps_doc_frac_chars_top_2gram | The fraction of characters in the top word 2gram. | Repetitiveness | [RefinedWeb](https://arxiv.org/abs/2306.01116), [Gopher](https://arxiv.org/abs/2112.11446) |
126
+ | rps_doc_frac_chars_top_3gram | The fraction of characters in the top word 3gram. | Repetitiveness | [RefinedWeb](https://arxiv.org/abs/2306.01116), [Gopher](https://arxiv.org/abs/2112.11446) |
127
+ | rps_doc_frac_chars_top_4gram | The fraction of characters in the top word 4gram. | Repetitiveness | [RefinedWeb](https://arxiv.org/abs/2306.01116), [Gopher](https://arxiv.org/abs/2112.11446) |
128
+ | rps_doc_ldnoobw_words | The number of sequences of words that are contained in the List-of-Dirty-Naughty-Obscene-and-Otherwise-Bad-Words blocklist. The blocklist is obtained from the [LDNOOBW](https://github.com/LDNOOBW/List-of-Dirty-Naughty-Obscene-and-Otherwise-Bad-Words) repo. | toxicity | [C4](https://arxiv.org/abs/1910.10683) |
129
+ | rps_doc_ut1_blacklist | A categorical id corresponding to the list of categories of the domain of the document. Categories are obtained from the UT1 blacklist. The list is obtained from [UT-Capitole](https://dsi.ut-capitole.fr/blacklists/). | toxicictiy | [RefinedWeb](https://arxiv.org/abs/2306.01116) |
130
+
131
+ #### Document and Token Counts for the Annotated (`head_middle`) part of the dataset
132
+
133
+ | | # Documents | Estimated Token count (deduped) |
134
+ |-------|-------------|---------------------------------|
135
+ | en | 24.5B | 20.5T |
136
+ | de | 2.7B | 3.0T |
137
+ | fr | 2.2B | 2.7T |
138
+ | es | 2.3B | 2.8T |
139
+ | it | 1.2B | 1.5T |
140
+ | Total | 32.9B | 30.4T |
141
 
142
  ### Languages
143
 
 
174
  ├── it_middle.minhash.parquet
175
  ```
176
 
177
+ Documents files, which contain the text, folow the schema defined by CCNet:
178
+
179
+ ```json
180
+ {
181
+ "url": "...",
182
+ "date_download": "2014-08-20T06:48:26Z",
183
+ "digest": "sha1:46OPKWZ7MAG5624VYYA3U3YH2MJ727B6",
184
+ "length": 1095,
185
+ "nlines": 8,
186
+ "source_domain": "170.48.11.165",
187
+ "title": "\ufeff",
188
+ "raw_content": "Dear ...",
189
+ "cc_segment": "crawl-data/CC-MAIN-2014-35/...",
190
+ "original_nlines": 11,
191
+ "original_length": 1174,
192
+ "line_ids": [0, 1, 3, 4, 6, 7, 8, 9],
193
+ "language": "en",
194
+ "language_score": 0.92,
195
+ "perplexity": 217.2,
196
+ "bucket": "head"
197
+ }
198
+ ```
199
+
200
+ The quality signals follow the schema
201
 
202
  ```json
203
  {
 
245
  ```
246
 
247
  where signal scores are encoded as a list of tuples `(start, end, score)`, where `start` and `end` are the locations in
248
+ the `raw_content` string where the `score` applies.
 
249
 
250
  ## Dataset Creation
251
 
252
+ The dataset is based on 84 snapshots provided by Common Crawl. Each snapshot was processed using the CCNet pipeline and
253
+ split into `head` `middle` `tail` buckets, depending on the perplexity score. In a second step, the documents in the
254
+ `head` and `middle` buckets were annotated with the quality signals described above. Finally, the documents were
255
+ deduplicated based on the text, using a Bloomfilter. The duplicates were kept in the dataset, but are marked in the
256
+ `duplicates` component.
257
 
258
  ## Citation
259
 
 
262
  ```
263
  @software{together2023redpajama-v2,
264
  author = {Together Computer},
265
+ title = {RedPajama-Data-v2: an Open Dataset for Training Large Language Models},
266
  month = October,
267
  year = 2023,
268
  url = {https://github.com/togethercomputer/RedPajama-Data}