---
title: Korean Toxicity Deobfuscation KOTOX
emoji: 🧼
colorFrom: purple
colorTo: indigo
sdk: gradio
sdk_version: 6.22.0
app_file: app.py
short_description: Deobfuscate and detoxify Korean text with KOTOX
python_version: "3.12"
startup_duration_timeout: 1h
---

# Korean Toxicity Deobfuscation (KOTOX)

This Space demonstrates the KOTOX model — a fine-tuned Llama-3-Korean-Bllossom-8B (LoRA adapter)
for Korean text deobfuscation and sanitization, from the paper
[Obfuscation Rules for Detecting and Detoxifying Korean Toxicity](https://arxiv.org/abs/2510.10961).

## Two modes

- **Deobfuscation** — restores obfuscated *neutral* text to clean, standard Korean.
- **Sanitization** — rewrites obfuscated *toxic* text into deobfuscated, non-toxic Korean
  while preserving the intended meaning.

## Model

- Base: [MLP-KTLim/llama-3-Korean-Bllossom-8B](https://huggingface.co/MLP-KTLim/llama-3-Korean-Bllossom-8B)
- LoRA adapter: [ssgyejin/llama-3-Korean-Bllossom-8B-deobfuscation-KOTOX](https://huggingface.co/ssgyejin/llama-3-Korean-Bllossom-8B-deobfuscation-KOTOX)
- Dataset: [ssgyejin/KOTOX](https://huggingface.co/datasets/ssgyejin/KOTOX)