Update README.md
Browse files
README.md
CHANGED
@@ -7,4 +7,41 @@ language:
|
|
7 |
base_model:
|
8 |
- answerdotai/ModernBERT-base
|
9 |
pipeline_tag: text-classification
|
10 |
-
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
7 |
base_model:
|
8 |
- answerdotai/ModernBERT-base
|
9 |
pipeline_tag: text-classification
|
10 |
+
---
|
11 |
+
|
12 |
+
# Machine-generated text detection prevents language model collapse
|
13 |
+
|
14 |
+
This model is part of the research presented in the paper [Machine-generated text detection prevents language model collapse](https://arxiv.org/abs/2502.15654), which proposes an approach to prevent model collapse based on importance sampling from a machine-generated text detector. The official implementation and training scripts are available in the GitHub repository: [GeorgeDrayson/model_collapse](https://github.com/GeorgeDrayson/model_collapse)
|
15 |
+
|
16 |
+
## Usage
|
17 |
+
|
18 |
+
To use the model for detecting machine-generated text:
|
19 |
+
|
20 |
+
|
21 |
+
```python
|
22 |
+
from transformers import AutoTokenizer, AutoModelForSequenceClassification
|
23 |
+
import torch
|
24 |
+
|
25 |
+
tokenizer = AutoTokenizer.from_pretrained("GeorgeDrayson/modernbert-mage")
|
26 |
+
model = AutoModelForSequenceClassification.from_pretrained("GeorgeDrayson/modernbert-mage")
|
27 |
+
|
28 |
+
text = "Your input text here."
|
29 |
+
inputs = tokenizer(text, return_tensors="pt")
|
30 |
+
outputs = model(**inputs)
|
31 |
+
probabilities = torch.nn.functional.softmax(outputs.logits, dim=-1)
|
32 |
+
print(f"Probability of machine-generated text: {probabilities[0][1].item():.4f}")
|
33 |
+
```
|
34 |
+
|
35 |
+
## Citation
|
36 |
+
|
37 |
+
If you use this model or find the research helpful, please cite:
|
38 |
+
|
39 |
+
|
40 |
+
```bibtex
|
41 |
+
@article{drayson2025machine,
|
42 |
+
title={Machine-generated text detection prevents language model collapse},
|
43 |
+
author={Drayson, George and Yilmaz, Emine and Lampos, Vasileios},
|
44 |
+
journal={arXiv preprint arXiv:2502.15654},
|
45 |
+
year={2025}
|
46 |
+
}
|
47 |
+
```
|