Skip to content

Editing the Wiki

Irregularpedia is a community wiki - anyone with an IrregularChat account can contribute. This guide explains how to use the in-page editor to make changes.

  1. Login - Click the user icon in the header and sign in with your IrregularChat SSO account
  2. Navigate - Go to the page you want to edit
  3. Click Edit - Click the pencil icon (✏️) next to the page title
  4. Make Changes - Edit the markdown with live preview
  5. Save - Click Save to publish your changes

The in-page editor provides a split-pane view:

Left PaneRight Pane
Markdown sourceLive preview

As you type, the preview updates in real-time so you can see how your changes will look.

  • Cmd/Ctrl + S - Save (when in editor)
  • Escape - Close save dialog
# Heading 1
## Heading 2
### Heading 3
**bold text**
*italic text*
`inline code`
[Link text](https://example.com)
[Internal link](/ai-ml/claude-code)
- Bullet point
- Another point
- Nested point
1. Numbered item
2. Second item
```python
def hello():
print("Hello, world!")
```
| Header 1 | Header 2 |
|----------|----------|
| Cell 1 | Cell 2 |
| Cell 3 | Cell 4 |

Starlight supports special callout blocks:

:::note[Note Title]
This is informational content.
:::
:::tip[Tip]
Helpful advice goes here.
:::
:::caution[Warning]
Be careful about this.
:::
:::danger[Danger]
Critical warning!
:::

When adding factual claims, statistics, or information from external sources, cite your sources using footnote syntax. This creates Wikipedia-style numbered references with bidirectional links — readers can click a citation number to jump to the source, and click the back-arrow (↩) to return to where they were reading.

Place [^N] after the claim, where N is the reference number:

FlashAttention-2 provides 2-4x faster attention[^1]. PyTorch 2.2+ includes
native support[^2]. DeepSpeed reduces memory by up to 8x[^1].

This renders as superscript numbers: FlashAttention-2 provides 2-4x faster attention1. When a source is cited multiple times, the footnote section automatically generates multiple back-arrows (↩, ↩², ↩³) — one for each place it was referenced.

At the bottom of the page, under a ## References heading, define each footnote:

## References
[^1]: Author or Organization. [Title of Source](https://example.com). Publisher, Year.
"Key quote from the source that supports the claim."
[^2]: Another Author. [Another Source](https://example.com/doc). Publisher.
"Relevant quote or summary."

Follow this pattern for each reference:

[^N]: Who. [What](URL). Where/When. "Key quote."
FieldExample
WhoEleutherAI or Dao, T. et al.
What[lm-evaluation-harness](https://github.com/EleutherAI/lm-evaluation-harness)
Where/WhenGitHub or arXiv:2211.09110, 2022 or PyTorch Blog, 2025
Key quote"Over 60 standard academic benchmarks for LLMs."
  1. Every new factual claim needs at least 2 sources — this prevents single-source bias and ensures claims are well-supported
  2. Link to primary sources — prefer official docs, GitHub repos, and papers over blog summaries
  3. Include a key quote — helps readers evaluate relevance without clicking through
  4. Number sequentially[^1], [^2], [^3], etc. in order of first appearance
  5. Keep an “Additional Resources” section — for useful links that aren’t directly cited in the text
---
title: "My Wiki Page"
tags: ["example"]
---
# My Wiki Page
LoRA reduces memory requirements by 10-20x compared to full fine-tuning[^1].
The technique freezes original weights and trains small low-rank matrices[^2].
Multiple frameworks support LoRA, including Axolotl and torchtune[^3].
## References
[^1]: Introl. [Fine-Tuning Infrastructure: LoRA, QLoRA, and PEFT at Scale](https://introl.com/blog/...).
Introl Blog, 2025. "PEFT methods can reduce memory requirements by roughly 10-20x."
[^2]: Databricks. [Efficient Fine-Tuning with LoRA](https://www.databricks.com/blog/...).
Databricks Blog. "LoRA freezes the original model weights and injects small trainable
low-rank matrices."
[^3]: Spheron Network. [Axolotl vs Unsloth vs TorchTune](https://www.spheron.network/blog/...).
Spheron Blog, 2026.
### Additional Resources
- [HuggingFace PEFT Library](https://github.com/huggingface/peft)
- [PyTorch Fine-Tuning Tutorial](https://pytorch.org/tutorials/...)

For more on citation best practices, tools, and why citations matter, see Research Citations.

Every wiki page starts with frontmatter (metadata):

---
title: Your Page Title
description: Brief description for search engines
tags: ["tag1", "tag2"]
---
# Your content starts here
  • title - The page title (shown in browser tab and navigation)
  • description - SEO description (shown in search results)
  • tags - Array of tags for organization
  • sidebar - Custom sidebar configuration

If someone else edited the page while you were editing, you’ll see a conflict message. You have two options:

  1. Reload - Load the latest version (your changes will be lost)
  2. Copy your work - Before reloading, copy your changes to paste back in

You can bookmark pages for quick access:

  1. Login to your account
  2. Click the bookmark icon on any page
  3. Access your bookmarks from the user menu

Currently, new pages must be created via Git (GitHub or Forgejo). The in-page editor only supports editing existing pages.

To create a new page:

  1. Go to GitHub or Forgejo
  2. Navigate to src/content/docs/ and the appropriate subfolder
  3. Create a new .md file with frontmatter
  4. Submit a pull request