Skip to content

Dataset · akoksal

LongForm instructions for long text generation

akoksal/LongForm

English instruction dataset reconstructed from real texts in Wikipedia, C4, and Stack Exchange.

What's inside

Contains pairs of instructions and detailed answers in English, with a volume between 10,000 and 100,000 records. For creation, the authors took ready-made human texts from the C4 and Wikipedia corpora, used LLMs to generate reverse instructions, and supplemented the set with examples from Stack Exchange, WikiHow, and tasks for summarization, letter writing, and grammar correction.

What you can do with it

  • Fine-tuning models to generate long, structured, and natural texts
  • Training LoRA adapters for summarization, editing, letter writing, and table-based answering
Tasks
Table question answering, Summarization, Text generation, Question answering
Languages
English
Size
10K–100K rows
Downloads this month
713
Likes
61

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("akoksal/LongForm")