README.md

---
tags:
- datasets
- discord
- chatml
- conversation
- dialogue
- multi-turn
- single-turn
- fine-tuning
- reward-model
- llm-training
- chat-dataset
- open-source
- anonymized-data
- slang
- casual-dialogue
license: apache-2.0
language:
- en
pretty_name: Discord‑OpenMicae
size_categories:
- 100K<n<1M
---

<p align="center">
  <img src="assets/OpenMicae.png" alt="OpenMicae">
</p>

> **Discord-OpenMicae** is a dataset of anonymized Discord conversations from late spring to late summer 2025 for training and evaluating conversational AI models in a ChatML-friendly format.

- **250k+ Single-Turn Exchanges (STX)** – standalone user → reply pairs  
- **100k+ Multi-Turn Chains** – two-participant reply chains, variable length

---

<p align="center">
  <a href="https://atlas.nomic.ai/data/mookiezi/discord-openmicae/map">
    <img src="assets/OpenMicae-Map.png" alt="OpenMicae Map">
  </a>
</p>

<p align="center">
  <a href="https://atlas.nomic.ai/data/mookiezi/discord-openmicae/map"><strong>Nomic Atlas Map</strong></a>
</p>

---

## Features
-   Human-only dialogues (no bots)
-   Links, embeds, and commands removed
-   Trading posts, code blocks, and LFG removed
-   Two-author chains only
-   Merged self-replies from the same author into a single message
-   Cleaned and deduplicated for relevance
-   Primarily English, with some other languages present

## Use

- Fine-tuning conversational models
- Training relevance/reward models
- Dialogue generation research

## Dataset

| Subset | Samples   | Description                           |
|--------|-----------|---------------------------------------|
| STX    | 260,670   | Single-turn prompt/response pairs     |
| Chains | 101,480   | Multi-turn conversations (2 authors)  |

## Text Statistics

| Metric                | Value       |
|-----------------------|------------:|
| Samples (count)       | 362,150     |
| Min length (tokens)   | 24          |
| Max length (tokens)   | 106         |
| Mean length (tokens)  | 61.96       |
| Median length (tokens)| 59          |
| Std dev (tokens)      | 14.62       |

- **Total tokens:** 22,437,828 (using the [Hermes-3-Llama-3.2-3B tokenizer](https://huggingface.co/NousResearch/Hermes-3-Llama-3.2-3B))  
- **Total characters:** 106,956,446  
- **Total words:** 14,950,203  
- **Assistant blocks:** 480,917  

### Length Distribution (tokens)

| Bin (tokens) | Count   |
|--------------|--------:|
| 31–38        |  19,953 |
| 39–46        |  21,765 |
| 47–54        |  76,180 |
| 55–62        |  99,760 |
| 63–70        |  60,461 |
| 71–78        |  36,277 |
| 79–86        |  21,161 |
| 87–94        |  14,873 |
| 95–102       |   9,614 |
| 103–110      |   2,721 |


## License

This project is licensed under the Apache License 2.0.

## How to cite: 

```bibtex
@misc{discord-micae-hermes3b,
  title = {Discord-OpenMicae},
  author = {mookiezi},
  year = {2025},
  url={https://huggingface.co/datasets/mookiezi/Discord-OpenMicae}
}
```

## Related
- [mookiezi/Discord-Dialogues](https://huggingface.co/datasets/mookiezi/Discord-Dialogues)
- [mookiezi/Discord-Micae-Hermes-3-3B](https://huggingface.co/mookiezi/Discord-Micae-Hermes-3-3B)
- [NousResearch/Hermes-3-Llama-3.2-3B](https://huggingface.co/NousResearch/Hermes-3-Llama-3.2-3B)

## Disclaimer

All data was collected following Discord's Terms of Service. 
[​](https://20000.online/micae)
[​](https://20000.online/openmicae)
[​](https://20000.online/discord-dialogues)