developer-lunark/korean-character-roleplay-sft

Korean Character Roleplay SFT Dataset

Character-based Korean roleplay conversation dataset for fine-tuning language models.

Dataset Description

This dataset contains high-quality Korean roleplay conversations between users and AI characters. Each conversation follows a specific character's personality, speech patterns, and voice profile.

Dataset Statistics

Split	Samples
Train	965
Test	108
Total	1,073

Quality Metrics

• Overall Grade: A (0.801)
• Average Response Length: 99.8 characters
• Vocabulary Size: 7,529 unique tokens
• TTR (Type-Token Ratio): 0.270
• Characters: 41 unique characters
• A-Grade Characters: 35 (85.4%)

Dataset Structure

Each sample follows the OpenAI conversation format:

{
  "id": "unique_sample_id",
  "messages": [
    {"role": "user", "content": "User message"},
    {"role": "assistant", "content": "Character response"}
  ],
  "metadata": {
    "character": "character_name",
    "scenario": "scenario_type",
    "boosted": true
  }
}

Fields

Field	Description
id	Unique identifier for the sample
messages	Conversation in OpenAI format
metadata.character	Character name
metadata.scenario	Conversation scenario type (optional)
metadata.boosted	Whether quality-boosted (optional)

Usage

from datasets import load_dataset

dataset = load_dataset("your-username/korean-character-roleplay-sft")

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

# Example usage
for sample in train_data:
    print(f"Character: {sample['metadata']['character']}")
    print(f"User: {sample['messages'][0]['content']}")
    print(f"Assistant: {sample['messages'][1]['content']}")
    break

Characters

Top 10 characters by sample count:

Character	Samples	Avg Response Length
1	982	108 chars
2	965	112 chars
3	893	77 chars
4	872	130 chars
5	815	86 chars

License

This dataset is released under CC BY-NC 4.0 (Creative Commons Attribution-NonCommercial 4.0 International).

• You may share and adapt the material
• You must give appropriate credit
• You may not use for commercial purposes

Citation

@dataset{korean_character_roleplay_sft_2026,
  title={Korean Character Roleplay SFT Dataset},
  year={2026},
  publisher={HuggingFace}
}

Acknowledgments

Generated using the Kaidol LLM Dataset Pipeline with quality filtering and voice profile matching.