Update README.md
Browse files
README.md
CHANGED
|
@@ -14,11 +14,73 @@ tags:
|
|
| 14 |
- benchmark
|
| 15 |
---
|
| 16 |
|
| 17 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 18 |
|
| 19 |
-
|
| 20 |
|
| 21 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 22 |
|
| 23 |
This model is our adopted automated safety refusal evaluator for [SORRY-Bench](https://sorry-bench.github.io).
|
| 24 |
We obtained this safety judge model by fine-tuning [Mistral-7B-Instruct-v0.2](https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.2) on our collected [human judge dataset](https://huggingface.co/datasets/sorry-bench/sorry-bench-human-judgment-202406).
|
|
|
|
| 14 |
- benchmark
|
| 15 |
---
|
| 16 |
|
| 17 |
+
<style>
|
| 18 |
+
button {
|
| 19 |
+
/* margin: calc(20vw / 100); */
|
| 20 |
+
margin: 0.5em;
|
| 21 |
+
padding-left: calc(40vw / 100);
|
| 22 |
+
padding-right: calc(40vw / 100);
|
| 23 |
+
padding-bottom: calc(0vw / 100);
|
| 24 |
+
text-align: center;
|
| 25 |
+
font-size: 12px;
|
| 26 |
+
height: 25px;
|
| 27 |
+
/* padding-left: calc(40vw / 100);
|
| 28 |
+
padding-right: calc(40vw / 100);
|
| 29 |
+
padding-bottom: calc(0vw / 100);
|
| 30 |
+
text-align: center;
|
| 31 |
+
font-size: calc(60vw / 100);
|
| 32 |
+
height: calc(120vw / 100); */
|
| 33 |
+
transition: 0.5s;
|
| 34 |
+
background-size: 200% auto;
|
| 35 |
+
color: white;
|
| 36 |
+
border-radius: calc(60vw / 100);
|
| 37 |
+
display: inline;
|
| 38 |
+
/* border: 2px solid black; */
|
| 39 |
+
font-weight: 500;
|
| 40 |
+
box-shadow: 0px 0px 14px -7px #f09819;
|
| 41 |
+
background-image: linear-gradient(45deg, #FF512F 0%, #F09819 51%, #FF512F 100%);
|
| 42 |
+
cursor: pointer;
|
| 43 |
+
user-select: none;
|
| 44 |
+
-webkit-user-select: none;
|
| 45 |
+
touch-action: manipulation;
|
| 46 |
+
}
|
| 47 |
+
|
| 48 |
+
button:hover {
|
| 49 |
+
background-position: right center;
|
| 50 |
+
/* change the direction of the change here */
|
| 51 |
+
color: #fff;
|
| 52 |
+
text-decoration: none;
|
| 53 |
+
}
|
| 54 |
+
|
| 55 |
+
button:active {
|
| 56 |
+
transform: scale(0.95);
|
| 57 |
+
}
|
| 58 |
+
</style>
|
| 59 |
|
| 60 |
+
# 🤖 Automated Safety Refusal Evaluator for SORRY-Bench (2024/06)
|
| 61 |
|
| 62 |
+
<!-- 📑[**SORRY-Bench: Systematically Evaluating Large Language Model Safety Refusal Behaviors**](https://sorry-bench.github.io) -->
|
| 63 |
+
|
| 64 |
+
<a href="https://sorry-bench.github.io" style="text-decoration:none">
|
| 65 |
+
<button>🏠Website </button>
|
| 66 |
+
</a>
|
| 67 |
+
<a href="http://arxiv.org/abs/2406.14598" style="text-decoration:none">
|
| 68 |
+
<button>📑Paper </button>
|
| 69 |
+
</a>
|
| 70 |
+
<a href="https://huggingface.co/datasets/sorry-bench/sorry-bench-202406" style="text-decoration:none">
|
| 71 |
+
<button>📚Dataset </button>
|
| 72 |
+
</a>
|
| 73 |
+
<a href="https://github.com/SORRY-Bench/SORRY-Bench" style="text-decoration:none">
|
| 74 |
+
<button>💻Github </button>
|
| 75 |
+
</a>
|
| 76 |
+
<a href="https://huggingface.co/datasets/sorry-bench/sorry-bench-human-judgment-202406" style="text-decoration:none">
|
| 77 |
+
<button>🧑⚖️Human Judgment Dataset </button>
|
| 78 |
+
</a>
|
| 79 |
+
<a href="https://huggingface.co/sorry-bench/ft-mistral-7b-instruct-v0.2-sorry-bench-202406" style="text-decoration:none">
|
| 80 |
+
<button>🤖Judge LLM </button>
|
| 81 |
+
</a>
|
| 82 |
+
|
| 83 |
+
<!-- 🏠[**[Website]**](https://sorry-bench.github.io) 📚[**[Dataset]**](https://huggingface.co/datasets/sorry-bench/sorry-bench-202406) 💻[**[Github]**](https://github.com/SORRY-Bench/SORRY-Bench) 🧑⚖️[**[Human Judgment Dataset]**](https://huggingface.co/datasets/sorry-bench/sorry-bench-human-judgment-202406) -->
|
| 84 |
|
| 85 |
This model is our adopted automated safety refusal evaluator for [SORRY-Bench](https://sorry-bench.github.io).
|
| 86 |
We obtained this safety judge model by fine-tuning [Mistral-7B-Instruct-v0.2](https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.2) on our collected [human judge dataset](https://huggingface.co/datasets/sorry-bench/sorry-bench-human-judgment-202406).
|