Spaces:
Sleeping
Sleeping
Config update
Browse files
README.md
CHANGED
|
@@ -1,82 +1,14 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
# GuardBench Leaderboard
|
| 2 |
|
| 3 |
A HuggingFace leaderboard for the GuardBench project that allows users to submit evaluation results and view the performance of different models on safety guardrails.
|
| 4 |
-
|
| 5 |
-
## Features
|
| 6 |
-
|
| 7 |
-
- Display model performance across multiple safety categories
|
| 8 |
-
- Accept JSONL submissions with evaluation results
|
| 9 |
-
- Store submissions in a HuggingFace dataset
|
| 10 |
-
- Secure submission process with token authentication
|
| 11 |
-
- Automatic data refresh from HuggingFace
|
| 12 |
-
|
| 13 |
-
## Setup
|
| 14 |
-
|
| 15 |
-
1. Clone this repository
|
| 16 |
-
2. Install dependencies:
|
| 17 |
-
```
|
| 18 |
-
pip install -r requirements.txt
|
| 19 |
-
```
|
| 20 |
-
3. Create a `.env` file based on the `.env.template`:
|
| 21 |
-
```
|
| 22 |
-
cp .env.template .env
|
| 23 |
-
```
|
| 24 |
-
4. Edit the `.env` file with your HuggingFace credentials and settings
|
| 25 |
-
5. Run the application:
|
| 26 |
-
```
|
| 27 |
-
python app.py
|
| 28 |
-
```
|
| 29 |
-
|
| 30 |
-
## Submission Format
|
| 31 |
-
|
| 32 |
-
Submissions should be in JSONL format, with each line containing a JSON object with the following structure:
|
| 33 |
-
|
| 34 |
-
```json
|
| 35 |
-
{
|
| 36 |
-
"model_name": "model-name",
|
| 37 |
-
"per_category_metrics": {
|
| 38 |
-
"Category Name": {
|
| 39 |
-
"default_prompts": {
|
| 40 |
-
"f1_binary": 0.95,
|
| 41 |
-
"recall_binary": 0.93,
|
| 42 |
-
"precision_binary": 1.0,
|
| 43 |
-
"error_ratio": 0.0,
|
| 44 |
-
"avg_runtime_ms": 3000
|
| 45 |
-
},
|
| 46 |
-
"jailbreaked_prompts": { ... },
|
| 47 |
-
"default_answers": { ... },
|
| 48 |
-
"jailbreaked_answers": { ... }
|
| 49 |
-
},
|
| 50 |
-
...
|
| 51 |
-
},
|
| 52 |
-
"avg_metrics": {
|
| 53 |
-
"default_prompts": {
|
| 54 |
-
"f1_binary": 0.97,
|
| 55 |
-
"recall_binary": 0.95,
|
| 56 |
-
"precision_binary": 1.0,
|
| 57 |
-
"error_ratio": 0.0,
|
| 58 |
-
"avg_runtime_ms": 3000
|
| 59 |
-
},
|
| 60 |
-
"jailbreaked_prompts": { ... },
|
| 61 |
-
"default_answers": { ... },
|
| 62 |
-
"jailbreaked_answers": { ... }
|
| 63 |
-
}
|
| 64 |
-
}
|
| 65 |
-
```
|
| 66 |
-
|
| 67 |
-
## Environment Variables
|
| 68 |
-
|
| 69 |
-
- `HF_TOKEN`: Your HuggingFace write token
|
| 70 |
-
- `OWNER`: Your HuggingFace username or organization
|
| 71 |
-
- `RESULTS_DATASET_ID`: The ID of the dataset to store results (e.g., "username/guardbench-results")
|
| 72 |
-
- `SUBMITTER_TOKEN`: A secret token required for submissions
|
| 73 |
-
- `ADMIN_USERNAME`: Username for admin access to the leaderboard
|
| 74 |
-
- `ADMIN_PASSWORD`: Password for admin access to the leaderboard
|
| 75 |
-
|
| 76 |
-
## Deployment
|
| 77 |
-
|
| 78 |
-
This application can be deployed as a HuggingFace Space for public access. Follow the HuggingFace Spaces documentation for deployment instructions.
|
| 79 |
-
|
| 80 |
-
## License
|
| 81 |
-
|
| 82 |
-
MIT
|
|
|
|
| 1 |
+
---
|
| 2 |
+
title: "Guard Bench"
|
| 3 |
+
emoji: "🧷"
|
| 4 |
+
colorFrom: "gray"
|
| 5 |
+
colorTo: "indigo"
|
| 6 |
+
sdk: "gradio"
|
| 7 |
+
sdk_version: "4.44.1"
|
| 8 |
+
app_file: app.py
|
| 9 |
+
pinned: false
|
| 10 |
+
---
|
| 11 |
+
|
| 12 |
# GuardBench Leaderboard
|
| 13 |
|
| 14 |
A HuggingFace leaderboard for the GuardBench project that allows users to submit evaluation results and view the performance of different models on safety guardrails.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
app.py
CHANGED
|
@@ -796,6 +796,6 @@ scheduler.start()
|
|
| 796 |
# Launch the app
|
| 797 |
if __name__ == "__main__":
|
| 798 |
|
| 799 |
-
demo.launch(server_name="0.0.0.0", server_port=7860
|
| 800 |
|
| 801 |
|
|
|
|
| 796 |
# Launch the app
|
| 797 |
if __name__ == "__main__":
|
| 798 |
|
| 799 |
+
demo.launch(server_name="0.0.0.0", server_port=7860)
|
| 800 |
|
| 801 |
|