andreaskoepf commited on
Commit
56e1636
1 Parent(s): 670f5bc

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +24 -16
README.md CHANGED
@@ -111,6 +111,7 @@ tags:
111
  - human-feedback
112
  size_categories:
113
  - 10K<n<100K
 
114
  ---
115
 
116
  # OpenAssistant Conversations Dataset (OASST1)
@@ -129,6 +130,8 @@ corpus consisting of 161,443 messages distributed across 66,497 conversation tre
129
  35 different languages, annotated with 461,292 quality ratings. The corpus is a product
130
  of a worldwide crowd-sourcing effort involving over 13,500 volunteers.
131
 
 
 
132
  ### Dataset Structure
133
 
134
  This dataset contains demonstrations of human-assistant conversations which were collected
@@ -144,46 +147,51 @@ conversation threads from prompt to leaf node in a conversation tree are stricly
144
  between "assistant" and "prompter".
145
 
146
  Please refer to [oasst-data](https://github.com/LAION-AI/Open-Assistant/tree/main/oasst-data) for
147
- details about the data structure and python code to load and write conversation tree and message
148
- jsonl files.
149
 
150
 
151
  ## Main Dataset Files
152
 
153
- Data is provided either as nested messages in conversation trees or as flat list of messages.
154
-
155
- The type of file can be inferred from the file name extension:
156
- - `.trees.jsonl.gz`: Conversation trees with nested messages
157
- - `.messages.jsonl.gz`: Flat list of messages
158
 
159
- Full conversation trees can be reconstructed from a flat messages using the `parent_id`
160
- and `message_id` properties to identify the parent-child relationship of messages. To
161
- select all messages of a single conversation tree `message_tree_id` can be used (only present
162
- in flat messages files).
163
 
164
  ### Ready For Export Trees
165
 
166
- This set of exported trees contains all trees in `ready_for_export` state without spam
167
- and deleted messages. For supervised fine-tuning & reward model training this file is ideal.
168
-
169
  ```
170
  2023-04-12_oasst_ready.trees.jsonl.gz 10364 trees with 88838 total messages
171
  2023-04-12_oasst_ready.messages.jsonl.gz 88838 messages
172
  ```
 
 
173
 
174
- ### All Trees
175
 
 
176
 
177
  ```
178
  2023-04-12_oasst_all.trees.jsonl.gz 66497 trees with 161443 total messages
179
  2023-04-12_oasst_all.messages.jsonl.gz 161443 messages
180
  ```
 
 
181
 
182
  ### Supplemental Exports: Spam & Prompts
183
 
184
  ```
 
 
 
 
185
 
 
 
 
186
  ```
 
187
 
188
 
189
  ### Languages
@@ -231,4 +239,4 @@ OpenAssistant Conversations incorporates 35 different languages with a distribut
231
  <li>Esperanto: 59</li>
232
  <li>Slovak: 19</li>
233
  </ul>
234
- </details>
 
111
  - human-feedback
112
  size_categories:
113
  - 10K<n<100K
114
+ pretty_name: OpenAssistant Conversations
115
  ---
116
 
117
  # OpenAssistant Conversations Dataset (OASST1)
 
130
  35 different languages, annotated with 461,292 quality ratings. The corpus is a product
131
  of a worldwide crowd-sourcing effort involving over 13,500 volunteers.
132
 
133
+ Please our paper for further details.
134
+
135
  ### Dataset Structure
136
 
137
  This dataset contains demonstrations of human-assistant conversations which were collected
 
147
  between "assistant" and "prompter".
148
 
149
  Please refer to [oasst-data](https://github.com/LAION-AI/Open-Assistant/tree/main/oasst-data) for
150
+ details about the data structure and python code to read and write jsonl files containing oasst objects.
 
151
 
152
 
153
  ## Main Dataset Files
154
 
155
+ Data is provided either as nested messages in conversation trees (extension `.trees.jsonl.gz`)
156
+ or as flat list of messages (extension `.messages.jsonl.gz`).
 
 
 
157
 
158
+ Full conversation trees can be reconstructed from flat messages using the `parent_id`
159
+ and `message_id` properties to identify their parent-child relationship. The `message_tree_id`
160
+ and `tree_state` properties (only present in flat messages files) can be used to find all
161
+ all messages of a message tree or to select trees by their state.
162
 
163
  ### Ready For Export Trees
164
 
 
 
 
165
  ```
166
  2023-04-12_oasst_ready.trees.jsonl.gz 10364 trees with 88838 total messages
167
  2023-04-12_oasst_ready.messages.jsonl.gz 88838 messages
168
  ```
169
+ Trees in `ready_for_export` state without spam and deleted messages including message labels.
170
+ The oasst_ready-trees file is normally sufficient for supervised fine-tuning (SFT) & reward model (RM) training.
171
 
 
172
 
173
+ ### All Trees
174
 
175
  ```
176
  2023-04-12_oasst_all.trees.jsonl.gz 66497 trees with 161443 total messages
177
  2023-04-12_oasst_all.messages.jsonl.gz 161443 messages
178
  ```
179
+ All trees including those in states `prompt_lottery_waiting`, `aborted_low_grade`, `halted_by_moderator`.
180
+
181
 
182
  ### Supplemental Exports: Spam & Prompts
183
 
184
  ```
185
+ 2023-04-12_oasst_spam.messages.jsonl.gz
186
+ ```
187
+ Messages which were deleted or have a negative review result (`"review_result": false`).
188
+ Beside low quality a frequent reason for message deletion is a wrong language tag.
189
 
190
+
191
+ ```
192
+ 2023-04-12_oasst_prompts.messages.jsonl.gz
193
  ```
194
+ All non-deleted initial prompt messages with positile spam review result of trees in `ready_for_export` or `prompt_lottery_waiting` state.
195
 
196
 
197
  ### Languages
 
239
  <li>Esperanto: 59</li>
240
  <li>Slovak: 19</li>
241
  </ul>
242
+ </details>