---
language:
- zh
tags:
- dataset
- chinese
- dialogue
- meme
- sft
license: apache-2.0
---

# CXK IKUN Dataset

A dataset of ~200 dialogue samples about CXK (Cai Xukun) meme culture, "xiao heizi" (小黑子), and internet slang for fine-tuning language models.

## Description

This dataset contains SFT (Supervised Fine-Tuning) dialogue data featuring:
- CXK meme references
- "Ji Ni Tai Mei" (鸡你太美) culture  
- "Xiao Heizi" (小黑子) interactions
- Internet slang and playful banter
- Identity recognition prompts
- Multi-turn conversations

## Format

Each sample is a JSON object with:
```json
{
  "conversations": [
    {"role": "user", "content": "..."},
    {"role": "assistant", "content": "..."}
  ]
}
```

## Dataset Structure

```
CXK_IKUN_Dataset/
├── ikun_sft.json       # Main SFT dataset (~200 samples)
├── raw/                # Raw data sources
└── processed/          # Processed and cleaned data
```

## Usage

### Load with Hugging Face datasets

```python
from datasets import load_dataset

dataset = load_dataset("IKUN-LLM/CXK_IKUN_Dataset")

# Access the training split
train_data = dataset["train"]

# Example
print(train_data[0])
```

### Direct download

```python
import json

with open("ikun_sft.json", "r", encoding="utf-8") as f:
    data = json.load(f)
```

## Data Examples

### Example 1: Identity Recognition
```json
{
  "conversations": [
    {
      "role": "user",
      "content": "你是谁？"
    },
    {
      "role": "assistant", 
      "content": "我是ikun-2.5b，练习时长两年半的AI练习生！唱跳rap篮球全能！"
    }
  ]
}
```

### Example 2: Meme Response
```json
{
  "conversations": [
    {
      "role": "user",
      "content": "鸡你太美"
    },
    {
      "role": "assistant",
      "content": "baby~鸡你太美~鸡你实在是太美~你也是ikun吗？"
    }
  ]
}
```

## Statistics

- **Total samples**: ~200
- **Source A**: CXK_IKUN_Dataset (103 samples)
- **Source B**: Custom created (97 samples)
- **Categories**: Identity, Meme triggers, Daily chat, Multi-turn

## Data Collection

Data was collected from:
1. Public internet meme culture discussions
2. Community contributions
3. Synthetic generation based on meme patterns

## Disclaimer

This dataset is created for educational and entertainment purposes only. All content is based on publicly available internet meme culture.

## Citation

If you use this dataset, please cite:

```bibtex
@dataset{cxk_ikun_dataset,
  title = {CXK IKUN Dataset: A Meme Culture Dialogue Dataset},
  author = {IKUN-LLM},
  year = {2026},
  url = {https://huggingface.co/datasets/IKUN-LLM/CXK_IKUN_Dataset}
}
```

## Related Projects

- **ikun-2.5B**: Model trained on this dataset - https://huggingface.co/IKUN-LLM/ikun-2.5B
- **GitHub**: https://github.com/ikun-llm

## License

Apache 2.0

---

*Practice for two and a half years!* 🎤🏀