Wikinium

Wikinium is alive.

The Living Wikipedia Dataset

Wikinium is a continuously evolving Wikipedia dataset optimized for Large Language Model pretraining.

Unlike traditional Wikipedia datasets that are generated once and eventually become outdated, Wikinium is a living dataset. Articles are continuously collected, processed, updated, and published as Wikipedia itself evolves.

Knowledge changes every day.

Wikinium changes with it.

---

What is a Living Dataset?

Most Wikipedia datasets represent a snapshot in time.

Wikinium represents a process.

New articles are continuously added, existing articles are continuously refreshed, and improvements propagate automatically throughout the dataset.

As a result, Wikinium remains aligned with the ever-changing state of Wikipedia rather than a historical dump captured months or years ago.

Continuous Updates

Wikinium is updated automatically approximately every 10 minutes to 1 hour, depending on processing conditions.

Updates include:

• Newly created Wikipedia articles
• Existing article revisions
• Dataset growth through continuous acquisition
• Quality improvements from the processing pipeline

The dataset is therefore never truly "finished".

It is alive.

---

Powered by Wikinium-Maker

Wikinium is generated and maintained by Wikinium-Maker, an autonomous processing system hosted on Hugging Face Spaces.

Current public instances:

• https://huggingface.co/spaces/Rikunarita-ORG/Wikinium-Maker
• https://huggingface.co/spaces/Rikunarita-ORG/Wikinium-Maker-2nd

These systems continuously perform:

1. Wikipedia article acquisition
2. Content extraction
3. Advanced noise removal
4. High-fidelity Markdown transformation
5. LLM pretraining optimization
6. Continuous publication to Rikunarita-ORG/Wikinium

The entire pipeline operates with minimal human intervention.

---

Built for LLM Pretraining

Raw Wikipedia pages contain significant amounts of information that are irrelevant for language model learning:

• HTML boilerplate
• Navigation structures
• Interface elements
• Styling metadata
• Template artifacts
• Other structural noise

Wikinium-Maker transforms Wikipedia articles into clean Markdown specifically optimized for machine learning workloads.

Key Goals

• Preserve knowledge
• Remove noise
• Maintain structure
• Improve token efficiency
• Maximize training quality

The resulting Markdown representation is designed to be significantly more suitable for language model pretraining than raw HTML dumps.

---

Observe Wikinium While It Lives

One of the unique aspects of Wikinium is that its creation process is publicly observable.

You can watch the active generation pipeline in real time through the Wikinium-Maker Spaces:

• https://huggingface.co/spaces/Rikunarita-ORG/Wikinium-Maker
• https://huggingface.co/spaces/Rikunarita-ORG/Wikinium-Maker-2nd

These Spaces allow anyone to observe the dataset's ongoing evolution as articles are collected, processed, and prepared for publication.

Rather than downloading a static artifact, you can watch the dataset being created.

---

Pipeline Overview

Wikipedia

↓ Continuous Acquisition

Wikinium-Maker

↓ Noise Removal

↓ Markdown Transformation

↓ LLM Optimization

↓ Quality Processing

Rikunarita-ORG/Wikinium

↓ Continuous Updates

Living Dataset

---

Core Characteristics

Feature	Wikinium
Living Dataset	Yes
Continuous Growth	Yes
Existing Article Refresh	Yes
Automated Operation	Yes
Markdown Native	Yes
LLM Optimized	Yes
Public Processing Pipeline	Yes
Hugging Face Native	Yes

---

Philosophy

Traditional datasets preserve a moment.

Wikinium preserves a flow.

It is not merely a collection of Wikipedia articles.

It is a continuously evolving knowledge corpus built for the next generation of language models.

Generated by Wikinium-Maker.

Continuously evolving with Wikipedia.

Continuously growing with human knowledge.

@misc{wikinium,
  title={Wikinium},
  author={Rikunarita-ORG},
  year={2026},
  url={https://huggingface.co/datasets/Rikunarita-ORG/Wikinium},
  publisher={Hugging Face}
}