Datasets:

Modalities:
Tabular
Text
Formats:
parquet
ArXiv:
Libraries:
Datasets
pandas
License:
Andreas Koepf commited on
Commit
1e678c3
1 Parent(s): b8629ee

update README & prompt/spam file names

Browse files
2023-11-05_oasst2_prompts.jsonl.gz → 2023-11-05_oasst2_prompts.messages.jsonl.gz RENAMED
File without changes
2023-11-05_oasst2_spam.jsonl.gz → 2023-11-05_oasst2_spam.messages.jsonl.gz RENAMED
File without changes
README.md CHANGED
@@ -71,47 +71,157 @@ dataset_info:
71
  num_examples: 6599
72
  download_size: 66674129
73
  dataset_size: 166813577
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
74
  ---
75
 
 
76
 
77
- ### 2023-11-05_oasst2_all.trees.jsonl.gz
78
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
79
  ```
80
- Trees : 70642
81
- Messages : 208584
 
 
 
 
 
 
82
  Oldest message : 2023-01-16 20:24:26.211711+00:00
83
  Youngest message : 2023-11-05 10:24:44.484910+00:00
84
- Detoxify ratings : 156570
85
- Accepted messages: 189288
86
- Deleted messages : 5414
87
  Tree counts by state:
88
- - ready_for_export: 13854
89
- - prompt_lottery_waiting: 44550
90
- - halted_by_moderator: 3089
91
- - initial_prompt_review: 4319
92
- - growing: 3102
93
- - aborted_low_grade: 1708
94
  - ranking: 20
95
  Message counts by language:
96
- - en: 85115
97
- - es: 47513
98
- - ru: 15990
99
- - zh: 11205
100
- - de: 8398
101
- - fr: 5841
102
- - pt-BR: 4540
103
- - th: 3236
104
- - ca: 2586
105
- - it: 2144
106
- - ja: 1904
107
- - uk-UA: 1889
108
- - ko: 1635
109
- - pl: 1510
110
- - eo: 1405
111
- - nl: 1354
112
- - ar: 1274
113
- - vi: 1137
114
- - fi: 1098
115
  - eu: 995
116
  - hu: 961
117
  - tr: 803
@@ -136,33 +246,43 @@ Message counts by language:
136
  - lt: 7
137
  ```
138
 
139
- ### Ready For Export Trees
140
 
141
  ```
142
- 2023-11-05_oasst2_ready.trees.jsonl.gz 13,854 trees with 135,174 total messages
143
- 2023-11-05_oasst2_ready.messages.jsonl.gz 135,174 messages
144
  ```
145
 
146
- Trees in ready_for_export state without spam and deleted messages including message labels. The oasst_ready-trees file usually is sufficient for supervised fine-tuning (SFT) & reward model (RM) training.
147
-
148
- ### All Trees
149
 
150
  ```
151
- 2023-11-05_oasst2_all.trees.jsonl.gz 70,642 trees with 208,584 total messages
152
- 2023-11-05_oasst2_all.messages.jsonl.gz 208,584 messages
153
  ```
154
 
155
- All trees, including those in states prompt_lottery_waiting (trees that consist of only one message, namely the initial prompt), aborted_low_grade (trees that stopped growing because the messages had low quality), and halted_by_moderator.
156
 
 
157
 
158
- ### Supplemental Exports: Spam & Prompts
 
 
159
 
160
- ```
161
- 2023-11-05_oasst2_spam.jsonl.gz 19,296 matching messages
 
 
 
 
 
162
  ```
163
 
164
- These are messages which were deleted or have a negative review result ("review_result": false). Besides low quality, a frequent reason for message deletion is a wrong language tag.
165
 
166
- ```
167
- 2023-11-05_oasst2_prompts.jsonl.gz 64,592 matching messages
168
- ```
 
 
 
 
 
 
 
71
  num_examples: 6599
72
  download_size: 66674129
73
  dataset_size: 166813577
74
+ language:
75
+ - en
76
+ - es
77
+ - ru
78
+ - de
79
+ - pl
80
+ - th
81
+ - vi
82
+ - sv
83
+ - bn
84
+ - da
85
+ - he
86
+ - it
87
+ - fa
88
+ - sk
89
+ - id
90
+ - nb
91
+ - el
92
+ - nl
93
+ - hu
94
+ - eu
95
+ - zh
96
+ - eo
97
+ - ja
98
+ - ca
99
+ - cs
100
+ - bg
101
+ - fi
102
+ - pt
103
+ - tr
104
+ - ro
105
+ - ar
106
+ - uk
107
+ - gl
108
+ - fr
109
+ - ko
110
+ tags:
111
+ - human-feedback
112
+ size_categories:
113
+ - 100K<n<1M
114
+ pretty_name: OpenAssistant Conversations Release 2
115
  ---
116
 
117
+ # Open Assistant Conversations Dataset Release 2 (OASST2)
118
 
119
+ ## Dataset Description
120
 
121
+ - **Homepage:** https://www.open-assistant.io/
122
+ - **Repository:** https://github.com/LAION-AI/Open-Assistant
123
+ - **Paper:** https://arxiv.org/abs/2304.07327
124
+
125
+
126
+
127
+ ### Ready For Export Trees
128
+
129
+ ```
130
+ 2023-11-05_oasst2_ready.trees.jsonl.gz 13,854 trees with 135,174 total messages
131
+ 2023-11-05_oasst2_ready.messages.jsonl.gz 135,174 messages
132
+ ```
133
+
134
+ #### 2023-11-05_oasst2_ready.trees.jsonl.gz Stats
135
+ ```
136
+ Trees : 13,854
137
+ Messages : 135,174
138
+ Oldest message : 2023-01-16 20:24:26.211711+00:00
139
+ Youngest message : 2023-11-04 15:23:03.239343+00:00
140
+ Detoxify ratings : 111,448
141
+ Accepted messages: 129,517
142
+ Deleted messages : 4,376
143
+ Tree counts by state:
144
+ - ready_for_export: 13,854
145
+ Message counts by language:
146
+ - en: 64,513
147
+ - es: 28,199
148
+ - ru: 13,935
149
+ - zh: 8,615
150
+ - de: 6,145
151
+ - fr: 3,880
152
+ - pt-BR: 2,699
153
+ - th: 1,560
154
+ - ca: 1,283
155
+ - it: 943
156
+ - uk-UA: 845
157
+ - ja: 788
158
+ - pl: 435
159
+ - eo: 295
160
+ - eu: 274
161
+ - vi: 207
162
+ - fi: 138
163
+ - hu: 113
164
+ - ar: 80
165
+ - nl: 72
166
+ - da: 44
167
+ - tr: 37
168
+ - ko: 24
169
+ - he: 24
170
+ - id: 12
171
+ - cs: 12
172
+ - bn: 1
173
+ - sv: 1
174
+ ```
175
+
176
+ Trees in ready_for_export state without spam and deleted messages including message labels. The oasst_ready-trees file usually is sufficient for supervised fine-tuning (SFT) & reward model (RM) training.
177
+
178
+ ### All Trees
179
+
180
+ ```
181
+ 2023-11-05_oasst2_all.trees.jsonl.gz 70,642 trees with 208,584 total messages
182
+ 2023-11-05_oasst2_all.messages.jsonl.gz 208,584 messages
183
  ```
184
+
185
+ All trees, including those in states prompt_lottery_waiting (trees that consist of only one message, namely the initial prompt), aborted_low_grade (trees that stopped growing because the messages had low quality), and halted_by_moderator.
186
+
187
+ #### 2023-11-05_oasst2_all.trees.jsonl.gz Stats
188
+
189
+ ```
190
+ Trees : 70,642
191
+ Messages : 208,584
192
  Oldest message : 2023-01-16 20:24:26.211711+00:00
193
  Youngest message : 2023-11-05 10:24:44.484910+00:00
194
+ Detoxify ratings : 156,570
195
+ Accepted messages: 189,288
196
+ Deleted messages : 5,414
197
  Tree counts by state:
198
+ - ready_for_export: 13,854
199
+ - prompt_lottery_waiting: 44,550
200
+ - halted_by_moderator: 3,089
201
+ - initial_prompt_review: 4,319
202
+ - growing: 3,102
203
+ - aborted_low_grade: 1,708
204
  - ranking: 20
205
  Message counts by language:
206
+ - en: 85,115
207
+ - es: 47,513
208
+ - ru: 15,990
209
+ - zh: 11,205
210
+ - de: 8,398
211
+ - fr: 5,841
212
+ - pt-BR: 4,540
213
+ - th: 3,236
214
+ - ca: 2,586
215
+ - it: 2,144
216
+ - ja: 1,904
217
+ - uk-UA: 1,889
218
+ - ko: 1,635
219
+ - pl: 1,510
220
+ - eo: 1,405
221
+ - nl: 1,354
222
+ - ar: 1,274
223
+ - vi: 1,137
224
+ - fi: 1,098
225
  - eu: 995
226
  - hu: 961
227
  - tr: 803
 
246
  - lt: 7
247
  ```
248
 
249
+ ### Supplemental Exports: Spam & Prompts
250
 
251
  ```
252
+ 2023-11-05_oasst2_spam.messages.jsonl.gz 19,296 matching messages
 
253
  ```
254
 
255
+ These are messages which were deleted or have a negative review result ("review_result": false). Besides low quality, a frequent reason for message deletion is a wrong language tag.
 
 
256
 
257
  ```
258
+ 2023-11-05_oasst2_prompts.messages.jsonl.gz 64,592 matching messages
 
259
  ```
260
 
261
+ These are all the kept initial prompt messages with positive review result (no spam) of trees in `ready_for_export` or `prompt_lottery_waiting` state.
262
 
263
+ ### Using the Huggingface Datasets
264
 
265
+ While HF datasets is ideal for tabular datasets, it is not a natural fit for nested data structures like the OpenAssistant conversation trees.
266
+ Nevertheless, we make all messages which can also be found in the file `2023-04-12_oasst_ready.trees.jsonl.gz` available in parquet as train/validation splits.
267
+ These are directly loadable by [Huggingface Datasets](https://pypi.org/project/datasets/).
268
 
269
+ To load the oasst1 train & validation splits use:
270
+
271
+ ```python
272
+ from datasets import load_dataset
273
+ ds = load_dataset("OpenAssistant/oasst2")
274
+ train = ds['train'] # len(train)=128575 (95%)
275
+ val = ds['validation'] # len(val)=6599 (5%)
276
  ```
277
 
278
+ The messages appear in depth-first order of the message trees.
279
 
280
+ Full conversation trees can be reconstructed from the flat messages table by using the `parent_id`
281
+ and `message_id` properties to identify the parent-child relationship of messages. The `message_tree_id`
282
+ and `tree_state` properties (only present in flat messages files) can be used to find all messages of a message tree or to select trees by their state.
283
+
284
+ ## Contact
285
+
286
+ - Discord [Open Assistant Discord Server](https://ykilcher.com/open-assistant-discord)
287
+ - GitHub: [LAION-AI/Open-Assistant](https://github.com/LAION-AI/Open-Assistant)
288
+ - E-Mail: [[email protected]](mailto:[email protected])