File size: 2,871 Bytes
1087f34
02ed07c
1087f34
7989976
 
 
 
 
 
 
 
54fbc6a
7989976
 
 
 
 
 
 
 
 
 
 
 
 
 
ae21ade
7989976
 
54fbc6a
0dfb586
7989976
 
 
 
 
9510c3b
7989976
 
 
9510c3b
7989976
9510c3b
7989976
 
 
 
9510c3b
02ed07c
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
---
pipeline_tag: text-classification
---

<br>

# RADAR Model Card

## Model Details

RADAR-Vicuna-7B is an AI-text detector trained via adversrial 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.
- **License:** Non-commercial license
- **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 be noted 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 parphraser 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 the Section 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.