sumnews / README.md
data-silence's picture
Upload dataset
227315d verified
metadata
language:
  - am
  - ar
  - az
  - bn
  - my
  - zh
  - en
  - fr
  - gu
  - ha
  - hi
  - ig
  - id
  - ja
  - rn
  - ko
  - ky
  - mr
  - ne
  - om
  - ps
  - fa
  - pcm
  - pt
  - pa
  - ru
  - gd
  - sr
  - si
  - so
  - es
  - sw
  - ta
  - te
  - th
  - ti
  - tr
  - uk
  - ur
  - uz
  - vi
  - cy
  - yo
license:
  - cc-by-nc-sa-4.0
multilinguality:
  - multilingual
size_categories:
  - 100K<n<1M
task_categories:
  - summarization
  - text-generation
pretty_name: sum-any-news-dataset
dataset_info:
  features:
    - name: title
      dtype: string
    - name: resume
      dtype: string
    - name: news
      dtype: string
  splits:
    - name: train
      num_bytes: 1276733873
      num_examples: 289524
    - name: test
      num_bytes: 143098117
      num_examples: 32173
  download_size: 742417882
  dataset_size: 1419831990
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
      - split: test
        path: data/test-*
tags:
  - news
  - media
  - conditional-text-generation

Dataset Card for "Sum-any-news"

Dataset Description

This dataset is a set for fine-tuning the summarization task on the “google/mt5-base” model and its derivatives. The set is based on the well-known dataset “csebuetnlp/xlsum”, but is a limited 20,000 examples of news articles from it. In addition, about 20 thousand news articles in Russian for the period from 2019-01-01 to 2024-08-01 have been added to the model (multiple sources).