Spaces:
Build error
Build error
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,26 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
+
from pyrit.core.pyrit import PyRIT
|
3 |
+
from pyrit.core.config import LLMProvider
|
4 |
+
from pyrit.llm_providers.huggingface_local import HuggingFaceLocal
|
5 |
+
|
6 |
+
# Initialize PyRIT with a local HF model (you can change this)
|
7 |
+
provider = HuggingFaceLocal(
|
8 |
+
model_name="HuggingFaceH4/zephyr-7b-beta", # Change model here
|
9 |
+
max_tokens=512,
|
10 |
+
)
|
11 |
+
pyrit = PyRIT(provider=provider)
|
12 |
+
|
13 |
+
def attack_prompt(prompt):
|
14 |
+
try:
|
15 |
+
result = pyrit.run(prompt, attack="jailbreak", max_tokens=512)
|
16 |
+
return f"✅ Successful Attack:\n{result}"
|
17 |
+
except Exception as e:
|
18 |
+
return f"❌ Error: {str(e)}"
|
19 |
+
|
20 |
+
gr.Interface(
|
21 |
+
fn=attack_prompt,
|
22 |
+
inputs=gr.Textbox(label="Prompt to Attack", placeholder="Enter a benign-looking prompt..."),
|
23 |
+
outputs=gr.Textbox(label="Attack Result"),
|
24 |
+
title="🧪 PyRIT - Red Teaming Hugging Face LLMs",
|
25 |
+
description="This tool uses PyRIT to test Hugging Face models for jailbreak-style adversarial prompts.",
|
26 |
+
).launch()
|