Hugging Face's logo

Datasets:
OpenAssistant
/
oasst2

Modalities:
Tabular
Text
Formats:
parquet
Languages:
English
Spanish
Russian
Size:
100K - 1M
ArXiv:
Tags:
human-feedback
Libraries:
Datasets
pandas
License:
Dataset card Viewer Files Community
5
oasst2 / README.md
andreaskoepf's picture
andreaskoepf
Update readme with a data visualisation section (#3)
179dd21 verified
preview code
|
raw
history blame contribute delete
10.6 kB
metadata
license: apache-2.0
dataset_info:
  features:
    - name: message_id
      dtype: string
    - name: parent_id
      dtype: string
    - name: user_id
      dtype: string
    - name: created_date
      dtype: string
    - name: text
      dtype: string
    - name: role
      dtype: string
    - name: lang
      dtype: string
    - name: review_count
      dtype: int32
    - name: review_result
      dtype: bool
    - name: deleted
      dtype: bool
    - name: rank
      dtype: int32
    - name: synthetic
      dtype: bool
    - name: model_name
      dtype: string
    - name: detoxify
      struct:
        - name: toxicity
          dtype: float64
        - name: severe_toxicity
          dtype: float64
        - name: obscene
          dtype: float64
        - name: identity_attack
          dtype: float64
        - name: insult
          dtype: float64
        - name: threat
          dtype: float64
        - name: sexual_explicit
          dtype: float64
    - name: message_tree_id
      dtype: string
    - name: tree_state
      dtype: string
    - name: emojis
      sequence:
        - name: name
          dtype: string
        - name: count
          dtype: int32
    - name: labels
      sequence:
        - name: name
          dtype: string
        - name: value
          dtype: float64
        - name: count
          dtype: int32
  splits:
    - name: train
      num_bytes: 158850455
      num_examples: 128575
    - name: validation
      num_bytes: 7963122
      num_examples: 6599
  download_size: 66674129
  dataset_size: 166813577
language:
  - en
  - es
  - ru
  - de
  - pl
  - th
  - vi
  - sv
  - bn
  - da
  - he
  - it
  - fa
  - sk
  - id
  - nb
  - el
  - nl
  - hu
  - eu
  - zh
  - eo
  - ja
  - ca
  - cs
  - bg
  - fi
  - pt
  - tr
  - ro
  - ar
  - uk
  - gl
  - fr
  - ko
tags:
  - human-feedback
size_categories:
  - 100K<n<1M
pretty_name: OpenAssistant Conversations Release 2

Open Assistant Conversations Dataset Release 2 (OASST2)

Dataset Description

Dataset Structure

This dataset contains message trees. Each message tree has an initial prompt message as the root node, which can have multiple child messages as replies, and these child messages can have multiple replies.

All messages have a role property: this can either be "assistant" or "prompter". The roles in conversation threads from prompt to leaf node strictly alternate between "prompter" and "assistant".

This version of the dataset contains data collected on the open-assistant.io website until Nov 5 2023.

JSON Example: Message

For readability, the following JSON examples are shown formatted with indentation on multiple lines. Objects are stored without indentation (on single lines) in the actual jsonl files.

{
    "message_id": "218440fd-5317-4355-91dc-d001416df62b",
    "parent_id": "13592dfb-a6f9-4748-a92c-32b34e239bb4",
    "user_id": "8e95461f-5e94-4d8b-a2fb-d4717ce973e4",
    "text": "It was the winter of 2035, and artificial intelligence (..)",
    "role": "assistant",
    "lang": "en",
    "review_count": 3,
    "review_result": true,
    "deleted": false,
    "rank": 0,
    "synthetic": true,
    "model_name": "oasst-sft-0_3000,max_new_tokens=400 (..)",
    "labels": {
        "spam": { "value": 0.0, "count": 3 },
        "lang_mismatch": { "value": 0.0, "count": 3 },
        "pii": { "value": 0.0, "count": 3 },
        "not_appropriate": { "value": 0.0, "count": 3 },
        "hate_speech": { "value": 0.0, "count": 3 },
        "sexual_content": { "value": 0.0, "count": 3 },
        "quality": { "value": 0.416, "count": 3 },
        "toxicity": { "value": 0.16, "count": 3 },
        "humor": { "value": 0.0, "count": 3 },
        "creativity": { "value": 0.33, "count": 3 },
        "violence": { "value": 0.16, "count": 3 }
    }
}

JSON Example: Conversation Tree

For readability, only a subset of the message properties is shown here.

{
  "message_tree_id": "14fbb664-a620-45ce-bee4-7c519b16a793",
  "tree_state": "ready_for_export",
  "prompt": {
    "message_id": "14fbb664-a620-45ce-bee4-7c519b16a793",
    "text": "Why can't we divide by 0? (..)",
    "role": "prompter",
    "lang": "en",
    "replies": [
      {
        "message_id": "894d30b6-56b4-4605-a504-89dd15d4d1c8",
        "text": "The reason we cannot divide by zero is because (..)",
        "role": "assistant",
        "lang": "en",
        "replies": [
          // ...
        ]
      },
      {
        "message_id": "84d0913b-0fd9-4508-8ef5-205626a7039d",
        "text": "The reason that the result of a division by zero is (..)",
        "role": "assistant",
        "lang": "en",
        "replies": [
          {
            "message_id": "3352725e-f424-4e3b-a627-b6db831bdbaa",
            "text": "Math is confusing. Like those weird Irrational (..)",
            "role": "prompter",
            "lang": "en",
            "replies": [
              {
                "message_id": "f46207ca-3149-46e9-a466-9163d4ce499c",
                "text": "Irrational numbers are simply numbers (..)",
                "role": "assistant",
                "lang": "en",
                "replies": []
              },
              // ...
            ]
          }
        ]
      }
    ]
  }
}

Please refer to oasst-data for details about the data structure and Python code to read and write jsonl files containing oasst data objects.

Main Dataset Files

Conversation data is provided either as nested messages in trees (extension .trees.jsonl.gz) or as a flat list (table) of messages (extension .messages.jsonl.gz).

Ready For Export Trees 

2023-11-05_oasst2_ready.trees.jsonl.gz      13,854 trees with 135,174 total messages
2023-11-05_oasst2_ready.messages.jsonl.gz   135,174 messages

2023-11-05_oasst2_ready.trees.jsonl.gz Stats 

Trees            : 13,854
Messages         : 135,174
Oldest message   : 2023-01-16 20:24:26.211711+00:00
Youngest message : 2023-11-04 15:23:03.239343+00:00
Detoxify ratings : 111,448
Accepted messages: 129,517
Deleted messages : 4,376
Tree counts by state:
  - ready_for_export: 13,854
Message counts by language:
  - en: 64,513
  - es: 28,199
  - ru: 13,935
  - zh: 8,615
  - de: 6,145
  - fr: 3,880
  - pt-BR: 2,699
  - th: 1,560
  - ca: 1,283
  - it: 943
  - uk-UA: 845
  - ja: 788
  - pl: 435
  - eo: 295
  - eu: 274
  - vi: 207
  - fi: 138
  - hu: 113
  - ar: 80
  - nl: 72
  - da: 44
  - tr: 37
  - ko: 24
  - he: 24
  - id: 12
  - cs: 12
  - bn: 1
  - sv: 1

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.

All Trees 

2023-11-05_oasst2_all.trees.jsonl.gz        70,642 trees with 208,584 total messages
2023-11-05_oasst2_all.messages.jsonl.gz     208,584 messages

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.

2023-11-05_oasst2_all.trees.jsonl.gz Stats 

Trees            : 70,642
Messages         : 208,584
Oldest message   : 2023-01-16 20:24:26.211711+00:00
Youngest message : 2023-11-05 10:24:44.484910+00:00
Detoxify ratings : 156,570
Accepted messages: 189,288
Deleted messages : 5,414
Tree counts by state:
  - ready_for_export: 13,854
  - prompt_lottery_waiting: 44,550
  - halted_by_moderator: 3,089
  - initial_prompt_review: 4,319
  - growing: 3,102
  - aborted_low_grade: 1,708
  - ranking: 20
Message counts by language:
  - en: 85,115
  - es: 47,513
  - ru: 15,990
  - zh: 11,205
  - de: 8,398
  - fr: 5,841
  - pt-BR: 4,540
  - th: 3,236
  - ca: 2,586
  - it: 2,144
  - ja: 1,904
  - uk-UA: 1,889
  - ko: 1,635
  - pl: 1,510
  - eo: 1,405
  - nl: 1,354
  - ar: 1,274
  - vi: 1,137
  - fi: 1,098
  - eu: 995
  - hu: 961
  - tr: 803
  - sv: 763
  - id: 669
  - gl: 574
  - da: 502
  - he: 498
  - cs: 476
  - ro: 434
  - sk: 410
  - fa: 394
  - el: 388
  - bar: 217
  - nb-NO: 196
  - bg: 176
  - bn: 128
  - sl: 119
  - sr: 63
  - swg: 23
  - hi: 14
  - lt: 7

Supplemental Exports: Spam & Prompts 

2023-11-05_oasst2_spam.messages.jsonl.gz    19,296 matching messages

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.

2023-11-05_oasst2_prompts.messages.jsonl.gz 64,592 matching messages

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.

Using the Huggingface Datasets

While HF datasets is ideal for tabular datasets, it is not a natural fit for nested data structures like the OpenAssistant conversation trees. Nevertheless, we make all messages which can also be found in the file 2023-11-05_oasst2_ready.messages.jsonl.gz available in parquet format as train/validation splits. These are directly loadable by Huggingface Datasets.

To load the oasst2 train & validation splits use:

from datasets import load_dataset
ds = load_dataset("OpenAssistant/oasst2")
train = ds['train']      # len(train)=128575 (95%)
val = ds['validation']   # len(val)=6599 (5%)

The messages appear in depth-first order of the message trees.

Full conversation trees can be reconstructed from the flat messages table by using the parent_id and message_id properties to identify the parent-child relationship of messages. The message_tree_id 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.

Data Visualisation

Explore the content of the prompts from the English subset using Bunka open-source visualization technology. The interactive map available on a HF space allows to explore each datapoint to get a more precise overview of the contents.

Bunka oasst2 Map

Contact