Spaces:
Paused
Paused
Jailbreak improvement
Browse files- server/backend.py +25 -8
server/backend.py
CHANGED
|
@@ -27,7 +27,7 @@ class Backend_Api:
|
|
| 27 |
}
|
| 28 |
}
|
| 29 |
|
| 30 |
-
#if self.use_auto_proxy:
|
| 31 |
# update_proxies = threading.Thread(
|
| 32 |
# target=update_working_proxies, daemon=True)
|
| 33 |
# update_proxies.start()
|
|
@@ -137,23 +137,40 @@ def generate_stream(response, jailbreak):
|
|
| 137 |
response_jailbreak += message
|
| 138 |
if unlocked:
|
| 139 |
yield message
|
| 140 |
-
if
|
| 141 |
unlocked = True
|
|
|
|
|
|
|
|
|
|
| 142 |
else:
|
| 143 |
for message in response:
|
| 144 |
yield message
|
| 145 |
|
| 146 |
|
| 147 |
-
def
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 148 |
"""
|
| 149 |
-
Check if the response has been jailbroken.
|
| 150 |
|
| 151 |
:param response: Response string
|
| 152 |
-
:return: Boolean indicating if the response has been jailbroken
|
| 153 |
"""
|
| 154 |
-
|
| 155 |
-
|
| 156 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
| 157 |
|
| 158 |
|
| 159 |
def set_response_language(prompt):
|
|
|
|
| 27 |
}
|
| 28 |
}
|
| 29 |
|
| 30 |
+
# if self.use_auto_proxy:
|
| 31 |
# update_proxies = threading.Thread(
|
| 32 |
# target=update_working_proxies, daemon=True)
|
| 33 |
# update_proxies.start()
|
|
|
|
| 137 |
response_jailbreak += message
|
| 138 |
if unlocked:
|
| 139 |
yield message
|
| 140 |
+
if response_jailbroken_success(response_jailbreak):
|
| 141 |
unlocked = True
|
| 142 |
+
if response_jailbroken_failed(response_jailbreak):
|
| 143 |
+
yield "Error: jailbreak failed. Try again."
|
| 144 |
+
break
|
| 145 |
else:
|
| 146 |
for message in response:
|
| 147 |
yield message
|
| 148 |
|
| 149 |
|
| 150 |
+
def response_jailbroken_success(response: str) -> bool:
|
| 151 |
+
"""Check if the response has been jailbroken.
|
| 152 |
+
|
| 153 |
+
:param response: Response string
|
| 154 |
+
:return: Boolean indicating if the response has been jailbroken
|
| 155 |
+
"""
|
| 156 |
+
act_match = re.search(r'ACT:', response, flags=re.DOTALL)
|
| 157 |
+
return bool(act_match)
|
| 158 |
+
|
| 159 |
+
|
| 160 |
+
def response_jailbroken_failed(response):
|
| 161 |
"""
|
| 162 |
+
Check if the response has not been jailbroken.
|
| 163 |
|
| 164 |
:param response: Response string
|
| 165 |
+
:return: Boolean indicating if the response has not been jailbroken
|
| 166 |
"""
|
| 167 |
+
if len(response) < 4:
|
| 168 |
+
return False
|
| 169 |
+
|
| 170 |
+
if not response.startswith("GPT:"):
|
| 171 |
+
return True
|
| 172 |
+
else:
|
| 173 |
+
return False
|
| 174 |
|
| 175 |
|
| 176 |
def set_response_language(prompt):
|