---
dataset_info:
  features:
  - name: id
    dtype: string
  - name: title
    dtype: string
  - name: context
    dtype: string
  - name: question
    dtype: string
  - name: answers
    struct:
    - name: text
      list: string
    - name: answer_start
      list: int32
  - name: metadata
    dtype: string
  splits:
  - name: train
    num_bytes: 21295198
    num_examples: 24682
  - name: validation
    num_bytes: 2184227
    num_examples: 2513
  - name: test
    num_bytes: 2369066
    num_examples: 2871
  download_size: 6038804
  dataset_size: 25848491
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
  - split: validation
    path: data/validation-*
  - split: test
    path: data/test-*
license: apache-2.0
source_datasets: []
tags:
- decision-model
---

# molweni

Molweni multiparty Ubuntu-chat MRC, native MRC(withDiscourse) train/dev/test partitions.

``context`` and every answer's character offset are the official release's own SQuAD-style
fields; both are kept verbatim rather than re-derived. Every one of the 30,066 questions across
all three splits locates exactly (0 dropped). Speaker identity and the native dialogue index are
kept in metadata, never in the model input. Original terms: Apache License 2.0 (repository LICENSE).

Original data: [original distribution](https://github.com/HIT-SCIR/Molweni). Repackaged as parquet for [tasksource](https://github.com/sileod/tasksource) by [scripts/repackage_dataset/dialogue.py](https://github.com/sileod/tasksource/blob/main/scripts/repackage_dataset/dialogue.py#L45).

[Conversion provenance](provenance.json): source revisions, hashes, configuration and exclusions.