---
license: cc-by-sa-4.0
language:
- en
pretty_name: GitHub Docs Corpus
size_categories:
- 1K<n<10K
task_categories:
- text-generation
tags:
- github
- documentation
- corpus
- sakthai
- house-of-sak
---

# GitHub Docs Corpus

A dataset containing **only information from GitHub** — the official
`github/docs` repository, i.e. the source of docs.github.com.

## Dataset Structure

- Files: `data/train.jsonl`
- Format: JSONL, one chunk per line
- Columns: `text` (cleaned doc chunk), `metadata` (`source`, `title`)
- Rows: 3,336

## Composition

- Source: `github/docs` (main branch), `content/` tree only — 3,734
  Markdown files covering GitHub features, workflows, webhooks, REST/GraphQL
  API docs, actions, security, and more.
- Processing: YAML frontmatter stripped (title kept), Liquid version tags
  (`{% ifversion %}`), images and HTML comments removed, markdown links
  flattened to their text, chunks split at heading boundaries (~2,000 chars).
- Excludes: files outside `content/`, assets, and pages with fewer than 40
  chars of body text.

## Loading

```python
from datasets import load_dataset

ds = load_dataset("Nanthasit/github-docs", split="train")
```