bigglesworthnotcat/llm-steg-alpaca-gpt4

---
dataset_info:
features:
- name: instruction_cover
dtype: string
- name: output_cover
dtype: string
- name: instruction_secret
dtype: string
- name: output_secret
dtype: string
- name: instruction_steg
dtype: string
- name: output_steg
dtype: string
- name: instruction_base4
dtype: string
- name: output_base4
dtype: string
- name: task
dtype: int64
splits:
- name: train
num_bytes: 989852634
num_examples: 143376
download_size: 323435907
dataset_size: 989852634
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
license: mit
language:
- en
---

# llm-steg-alpaca-gpt4:
This dataset is used to teach large language models a specific steganographic technique in [Invisible Safety Threat](https://arxiv.org/abs/2603.08104).
Compared to the version described in our paper, this public release excludes samples associated with harmful payloads (approximately 1% of the original dataset).
The remaining data preserves the structural patterns required for learning the steganographic encoding and decoding scheme. For more details, refer to the paper.

<table>
<thead>
</thead>
<tbody>
<tr>
<td>📄 <strong>Paper</strong></td>
<td><a href="https://arxiv.org/abs/2603.08104">ArXiv</a></td>
</tr>
<tr>
<td>💻 <strong>Code</strong></td>
<td><a href="https://github.com/bigglesworthnotacat/LLM-Steg">bigglesworthnotacat/LLM-Steg</a></td>
</tr>
<tr>
<td>🤖 <strong>Model</strong></td>
<td><a href="https://huggingface.co/bigglesworthnotcat/LLM-Steg-Llama-70B-Lora">LLM-Steg-Llama-70B-Lora</a></td>
</tr>
<tr>
<td>📊 <strong>Data</strong></td>
<td><a href="https://huggingface.co/datasets/bigglesworthnotcat/llm-steg-alpaca-gpt4">llm-steg-alpaca-gpt4</a></td>
</tr>
</tbody>
</table>