Update README.md
Browse files
README.md
CHANGED
@@ -1,3 +1,50 @@
|
|
1 |
---
|
2 |
license: mit
|
3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
---
|
2 |
license: mit
|
3 |
---
|
4 |
+
|
5 |
+
|
6 |
+
# π BookMIA Datasets
|
7 |
+
|
8 |
+
The **BookMIA datasets** serve as a benchmark designed to evaluate membership inference attack (MIA) methods, specifically in detecting pretraining data from OpenAI models that are released before 2023 (such as text-davinci-003).
|
9 |
+
The dataset contains non-member and member data: non-member data consists of text excerpts from books first published in 2023, while member data includes text excerpts from older books, as categorized by Chang et al. in 2023.
|
10 |
+
|
11 |
+
### π Applicability
|
12 |
+
|
13 |
+
The datasets can be applied to various OpenAI models released before **2023**:
|
14 |
+
|
15 |
+
- text-davinci-001
|
16 |
+
- text-davinci-002
|
17 |
+
- ... and more.
|
18 |
+
|
19 |
+
## Loading the datasets
|
20 |
+
|
21 |
+
To load the dataset:
|
22 |
+
|
23 |
+
```python
|
24 |
+
from datasets import load_dataset
|
25 |
+
|
26 |
+
LENGTH = 64
|
27 |
+
dataset = load_dataset("swj0419/BookMIA")
|
28 |
+
```
|
29 |
+
* Text Lengths: `512`.
|
30 |
+
* *Label 0*: Refers to the unseen data during pretraining. *Label 1*: Refers to the seen data.
|
31 |
+
|
32 |
+
## π οΈ Codebase
|
33 |
+
|
34 |
+
For evaluating MIA methods on our datasets, visit our [GitHub repository](https://github.com/swj0419/detect-pretrain-code).
|
35 |
+
|
36 |
+
## β Citing our Work
|
37 |
+
|
38 |
+
If you find our codebase and datasets beneficial, kindly cite our work:
|
39 |
+
|
40 |
+
```bibtex
|
41 |
+
@misc{shi2023detecting,
|
42 |
+
title={Detecting Pretraining Data from Large Language Models},
|
43 |
+
author={Weijia Shi and Anirudh Ajith and Mengzhou Xia and Yangsibo Huang and Daogao Liu and Terra Blevins and Danqi Chen and Luke Zettlemoyer},
|
44 |
+
year={2023},
|
45 |
+
eprint={2310.16789},
|
46 |
+
archivePrefix={arXiv},
|
47 |
+
primaryClass={cs.CL}
|
48 |
+
}
|
49 |
+
|
50 |
+
[1] Kent K Chang, Mackenzie Cramer, Sandeep Soni, and David Bamman. Speak, memory: An archaeology of books known to chatgpt/gpt-4. arXiv preprint arXiv:2305.00118, 2023.
|