|
--- |
|
pipeline_tag: text-classification |
|
--- |
|
|
|
<br> |
|
|
|
# RADAR Model Card |
|
|
|
## Model Details |
|
|
|
RADAR-Vicuna-7B is an AI-text detector trained via adversarial learning between the detector and a paraphraser on human-text corpus ([OpenWebText](https://huggingface.co/datasets/Skylion007/openwebtext)) and AI-text corpus generated |
|
based on [OpenWebText](https://huggingface.co/datasets/Skylion007/openwebtext). |
|
|
|
- **Developed by:** [TrustSafeAI](https://huggingface.co/TrustSafeAI) |
|
- **Model type:** An encoder-only language model based on the transformer architecture (RoBERTa). |
|
- **License:** [Non-commercial license](https://huggingface.co/lmsys/vicuna-7b-v1.1#model-details) (inherited from Vicuna-7B-v1.1) |
|
- **Trained from model:** [RoBERTa](https://arxiv.org/abs/1907.11692) |
|
|
|
|
|
### Model Sources |
|
|
|
- **Project Page:** https://radar.vizhub.ai/ |
|
- **Paper:** https://arxiv.org/abs/2307.03838 |
|
- **IBM Blog Post:** https://research.ibm.com/blog/AI-forensics-attribution |
|
|
|
## Uses |
|
Users could use this detector to assist them in detecting text generated by large language models. |
|
Please note that this detector is trained on AI-text generated by Vicuna-7B-v1.1. As the model only supports [non-commercial use](https://huggingface.co/lmsys/vicuna-7b-v1.1#model-details), the intended users are **not allowed to involve this detector into commercial activities**. |
|
|
|
## Get Started with the Model |
|
Please refer to the following guidelines to see how to locally run the downloaded model or use our API service hosted on Huggingface Space. |
|
- Google Colab Demo: https://colab.research.google.com/drive/1r7mLEfVynChUUgIfw1r4WZyh9b0QBQdo?usp=sharing |
|
- Huggingface API Documentation: https://trustsafeai-radar-ai-text-detector.hf.space/?view=api |
|
|
|
## Training Pipeline |
|
|
|
We propose adversarial learning between a paraphraser and our detector. The paraphraser's goal is to make the AI-generated text more like human-writen and the detector's goal is to |
|
promote it's ability to identify the AI-text. |
|
|
|
- **(Step 1) Training Data preparation**: Before training, we use Vicuna-7B to generate AI-text by performing text completion based on the prefix span of human-text in [OpenWebText](https://huggingface.co/datasets/Skylion007/openwebtext). |
|
|
|
- **(Step 2) Update the paraphraser** During training, the paraphraser will do paraphrasing on the AI-text generated in **Step 1**. And then collect the reward returned by the detector to update the paraphraser using Proxy Proximal Optimization loss. |
|
|
|
- **(Step 3) Update the detector** The detector is optimized using the logistic loss on the human-text, AI-text and paraphrased AI-text. |
|
|
|
See more details in Sections 3 and 4 of this [paper](https://arxiv.org/pdf/2307.03838.pdf). |
|
|
|
## Ethical Considerations |
|
We suggest users use our tool to assist with identifying AI-written content at scale and with discretion. If the detection result is to be used as evidence, further validation steps |
|
are necessary as RADAR cannot always make correct predictions. |
