Datasets:

ArXiv:
Maurice Weber commited on
Commit
84a48e2
β€’
1 Parent(s): 949be9c

update README.md

Browse files
Files changed (1) hide show
  1. README.md +118 -30
README.md CHANGED
@@ -12,40 +12,62 @@ pretty_name: Red Pajama V2 Data Foundation
12
 
13
  ### Getting Started
14
 
 
 
 
 
 
 
 
 
15
  ```python
16
  from datasets import load_dataset
17
 
18
- ds = load_dataset("togethercomputer/RedPajama-Data-V2", name="en-head-middle-all")
19
  ```
20
 
21
- Or you can directly download the files using the following command:
 
 
 
22
 
23
  ```bash
24
- wget 'https://data.together.xyz/redpajama-data-v2/v1.0.0/urls.txt'
25
- while read line; do
26
- dload_loc=${line#https://data.together.xyz/redpajama-data-v2/v1.0.0/}
27
- mkdir -p $(dirname $dload_loc)
28
- wget "$line" -O "$dload_loc"
29
- done < urls.txt
30
- ```
31
-
32
- After downloading the files, you can load the dataset from disk by setting the `RED_PAJAMA_DATA_DIR` environment
33
- variable XXXX TODO XXX
34
 
35
- A smaller sample of the dataset can be downloaded via
 
36
 
37
- ```python
38
- from datasets import load_dataset
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
39
 
40
- ds = load_dataset("togethercomputer/RedPajama-Data-V2", name="en-sample")
41
  ```
42
 
43
- A full set of scripts to recreate the dataset from can be
44
  found [here](https://github.com/togethercomputer/RedPajama-Data).
45
 
46
  ### Dataset Summary
47
 
48
- TODO: Write a sentence about the dataset
 
49
 
50
  ### Languages
51
 
@@ -53,25 +75,93 @@ English, German, French, Italian, Spanish
53
 
54
  ## Dataset Structure
55
 
56
- The dataset structure is as follows:
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
57
 
58
  ```json
59
  {
60
- TODO
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
61
  }
62
  ```
63
 
64
- ## Dataset Creation
 
65
 
66
- XXXX
67
 
68
- ### Commoncrawl
69
- TODO
70
 
71
- To cite RedPajama, please use:
72
 
73
  ```
74
- @software{together2023redpajama,
75
  author = {Together Computer},
76
  title = {RedPajama-Data-v2: a living data foundation for training open LLM models},
77
  month = October,
@@ -80,13 +170,11 @@ To cite RedPajama, please use:
80
  }
81
  ```
82
 
83
- ### License
84
-
85
- TODO: double check this
86
 
87
  Please refer to the licenses of the data subsets you use.
88
 
89
- * [Common Crawl Foundation Terms of Use](https://commoncrawl.org/terms-of-use/full/)
90
 
91
  <!--
92
  ### Annotations
 
12
 
13
  ### Getting Started
14
 
15
+ The full RedPajama-V2 dataset is a data foundation that includes a over 100B text documents coming from 84 CommonCrawl
16
+ snapshots and processed using the [CCNet](https://github.com/facebookresearch/cc_net) pipeline. Out of these, there are
17
+ 30B documents in the corpus that additionally come with quality signals.
18
+
19
+ Check out our [blog post](XXXXX) for more details on the build process, dataset structure and schema.
20
+
21
+ To familiarize yourself with the dataset, you can load the sample dataset with the following command:
22
+
23
  ```python
24
  from datasets import load_dataset
25
 
26
+ ds = load_dataset("togethercomputer/RedPajama-Data-V2", name="sample")
27
  ```
28
 
29
+ Alternatively, you can also directly download the files using the following instructions, using english data from the
30
+ `2023-06` snapshot and the `head_middle` partition as an example. The full set of CC snapshots included in the dataset
31
+ is given in `_CC_SNAPSHOT_IDS`, and the available partitions are `tail`, `head_middle`. The available language tags are
32
+ `en`, `de`, `fr`, `es`, `it`.
33
 
34
  ```bash
35
+ CC_SNAPSHOT="2023-06"
36
+ LANG="en"
37
+ PARTITION="head_middle"
38
+ BASE_URL="https://data.together.xyz/redpajama-data-v2/v1.0.0/"
 
 
 
 
 
 
39
 
40
+ listings_file="${LANG}-${CC_SNAPSHOT}-${PARTITION}.txt"
41
+ wget "${BASE_URL}/listings/${listings_file}"
42
 
43
+ # download documents
44
+ while read line; do
45
+ url="${BASE_URL}/documents/${line}.json.gz"
46
+ dest="documents/${line}.json.gz"
47
+ mkdir -p $(dirname $dest)
48
+ wget "$line" -O "$dest"
49
+ done <"${LANG}-${CC_SNAPSHOT}-${PARTITION}.txt"
50
+
51
+ # download other components
52
+ COMPS=("quality_signals" "minhash" "duplicates")
53
+ for comp in "${COMPS[@]}"; do
54
+ while read line; do
55
+ url="${BASE_URL}/${comp}/${line}.${comp}.json.gz"
56
+ dest="${comp}/${line}.${comp}.json.gz"
57
+ mkdir -p $(dirname $dest)
58
+ wget "$line" -O "$dest"
59
+ done <"${LANG}-${CC_SNAPSHOT}-${PARTITION}.txt"
60
+ done
61
 
 
62
  ```
63
 
64
+ A full set of scripts to recreate the dataset including the quality signals can be
65
  found [here](https://github.com/togethercomputer/RedPajama-Data).
66
 
67
  ### Dataset Summary
68
 
69
+ RedPajama-V2 is a data foundation for which includes over 100B text documents, out of which 30B documents come with
70
+ quality annotations.
71
 
72
  ### Languages
73
 
 
75
 
76
  ## Dataset Structure
77
 
78
+ The datset is structure into four components, each following the same key structure:
79
+
80
+ ```
81
+ β”œβ”€β”€ documents
82
+ β”œβ”€β”€ 2018-43
83
+ β”œβ”€β”€ 0000
84
+ β”œβ”€β”€ en_head.json.gz
85
+ β”œβ”€β”€ ...
86
+ β”œβ”€β”€ it_middle.json.gz
87
+ β”œβ”€β”€ quality_signals
88
+ β”œβ”€β”€ 2018-43
89
+ β”œβ”€β”€ 0000
90
+ β”œβ”€β”€ en_head.signals.json.gz
91
+ β”œβ”€β”€ ...
92
+ β”œβ”€β”€ it_middle.json.gz
93
+ β”œβ”€β”€ duplicates
94
+ β”œβ”€β”€ 2018-43
95
+ β”œβ”€β”€ 0000
96
+ β”œβ”€β”€ en_head.duplicates.parquet
97
+ β”œβ”€β”€ ...
98
+ β”œβ”€β”€ it_middle.duplicates.parquet
99
+ β”œβ”€β”€ minhash
100
+ β”œβ”€β”€ 2018-43
101
+ β”œβ”€β”€ 0000
102
+ β”œβ”€β”€ en_head.minhash.parquet
103
+ β”œβ”€β”€ ...
104
+ β”œβ”€β”€ it_middle.minhash.parquet
105
+ ```
106
+
107
+ Documents files, which contain the text, folow the schema defined by CCNet, and the quality signals follow the schema
108
 
109
  ```json
110
  {
111
+ "id": "2018-43/0000/en_head.json.gz/0",
112
+ "id_int": 7972430436813205988,
113
+ "metadata": {
114
+ "cc_segment": "crawl-data/...",
115
+ "cc_net_source": "2018-43/0000/en_head.json.gz",
116
+ "url": "...",
117
+ "source_domain": "...",
118
+ "language": "en",
119
+ "snapshot_id": "2018-43"
120
+ },
121
+ "quality_signals": {
122
+ "ccnet_original_length": [
123
+ [
124
+ 0,
125
+ 7033,
126
+ 8711.0
127
+ ]
128
+ ],
129
+ ...,
130
+ "rps_doc_stop_word_fraction": [
131
+ [
132
+ 0,
133
+ 7033,
134
+ 0.45121107
135
+ ]
136
+ ],
137
+ "rps_lines_num_words": [
138
+ [
139
+ 0,
140
+ 25,
141
+ 2
142
+ ],
143
+ ...,
144
+ [
145
+ 6980,
146
+ 7033,
147
+ 10
148
+ ]
149
+ ]
150
+ }
151
  }
152
  ```
153
 
154
+ where signal scores are encoded as list of tuple `(start, end, score)`, where `start` and `end` are the locations in the
155
+ `raw_content` string where the `score` applies.
156
 
157
+ ## Dataset Creation
158
 
159
+ The dataset is based on 84 snapshots provided by CommonCrawl.
 
160
 
161
+ To cite RedPajama-V2, please use:
162
 
163
  ```
164
+ @software{together2023redpajama-v2,
165
  author = {Together Computer},
166
  title = {RedPajama-Data-v2: a living data foundation for training open LLM models},
167
  month = October,
 
170
  }
171
  ```
172
 
173
+ ### License ---- TODO ----
 
 
174
 
175
  Please refer to the licenses of the data subsets you use.
176
 
177
+ * [Common Crawl Foundation Terms of Use](https://commoncrawl.org/terms-of-use)
178
 
179
  <!--
180
  ### Annotations