helenai's picture
Upload README.md
5f3d7b4

Summarize long texts

This dataset provides long prompts intended to use for testing language models with long inputs with different sizes.

Columns

  • id: numerical id
  • source: the source of the prompt text
  • length: indication of length of the prompt
  • text: the text of the prompt

Item lengths

Exact prompt length depend on the tokenizer for the model, and can differ quite a bit with different tokenizers. The lengths included in the dataset should be seen as relative/approximate.

id     source              length
---------------------------------
1      alice                 1000
2      sherlock              2000
3      lookingglass          3000
4      bluefairy             4000
5      alice                 5000
6      greatexpectations     6000
7      littlewomen           7000
8      timemachine           8000
9      sherlock              9000
10     lookingglass         10000
11     tomsawyer            12000
12     littlewomen          14000
13     windwillows          16000
14     sherlock             20000

Sources

All the texts are from public domain books:

alice             : https://ia801604.us.archive.org/6/items/alicesadventures19033gut/19033.txt
lookingglass      : https://www.gutenberg.org/files/12/12-0.txt
bluefairy         : https://www.gutenberg.org/files/503/503-0.txt
sherlock          : https://www.gutenberg.org/files/1661/1661-0.txt
greatexpectations : https://www.gutenberg.org/files/1400/1400-0.txt
littlewomen       : https://www.gutenberg.org/files/514/514-0.txt
timemachine       : https://www.gutenberg.org/files/35/35-0.txt
tomsawyer         : https://www.gutenberg.org/files/74/74-0.txt
windwillows       : https://www.gutenberg.org/files/289/289-0.txt