Update README.md
Browse files
README.md
CHANGED
@@ -1,3 +1,54 @@
|
|
1 |
---
|
2 |
license: mit
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
---
|
2 |
license: mit
|
3 |
+
language:
|
4 |
+
- en
|
5 |
+
tags:
|
6 |
+
- membership inference
|
7 |
+
- privacy
|
8 |
+
pretty_name: MIMIR
|
9 |
+
size_categories:
|
10 |
+
- 1K<n<10K
|
11 |
---
|
12 |
+
|
13 |
+
# MIMIR
|
14 |
+
|
15 |
+
These datasets serve as a benchmark designed to evaluate membership inference attack (MIA) methods, specifically in detecting pretraining data from extensive large language models.
|
16 |
+
|
17 |
+
## 📌 Applicability
|
18 |
+
|
19 |
+
The datasets can be applied to any model trained on The Pile, including (but not limited to):
|
20 |
+
- GPTNeo
|
21 |
+
- Pythia
|
22 |
+
- OPT
|
23 |
+
|
24 |
+
## Loading the datasets
|
25 |
+
To load the dataset:
|
26 |
+
|
27 |
+
```python
|
28 |
+
from datasets import load_dataset
|
29 |
+
|
30 |
+
dataset = load_dataset("iamgroot42/mimir", split=f"pile_cc")
|
31 |
+
```
|
32 |
+
|
33 |
+
- Available Splits: `arxiv`, `wikipedia_en`, `pile_cc`, `dm_mathematics`, `pubmed_central`, `full_pile`, ...
|
34 |
+
- Labels:
|
35 |
+
- _0_: Refers to the unseen data during pretraining.
|
36 |
+
- _1_: Refers to the seen data.
|
37 |
+
|
38 |
+
## 🛠️ Codebase
|
39 |
+
For evaluating MIA methods on our datasets, visit our [GitHub repository](http://github.com/iamgroot42/mimir).
|
40 |
+
|
41 |
+
## ⭐ Citing our Work
|
42 |
+
|
43 |
+
If you find our codebase and datasets beneficial, kindly cite our work:
|
44 |
+
|
45 |
+
```bibtex
|
46 |
+
@inproceedings{duan2024disentangling,
|
47 |
+
title={Disentangling Challenges in Membership Inference for Large Language Models},
|
48 |
+
author={},
|
49 |
+
booktitle={,
|
50 |
+
year={2024}
|
51 |
+
}
|
52 |
+
```
|
53 |
+
|
54 |
+
|