File size: 7,080 Bytes
e3278e4 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 |
# +-------------------------------------------------------------+
#
# Use AporiaAI for your LLM calls
#
# +-------------------------------------------------------------+
# Thank you users! We ❤️ you! - Krrish & Ishaan
import sys
import os
sys.path.insert(
0, os.path.abspath("../..")
) # Adds the parent directory to the system path
from typing import Optional, Literal, Any
import litellm
import sys
from litellm.proxy._types import UserAPIKeyAuth
from litellm.integrations.custom_guardrail import CustomGuardrail
from fastapi import HTTPException
from litellm._logging import verbose_proxy_logger
from litellm.proxy.guardrails.guardrail_helpers import should_proceed_based_on_metadata
from litellm.litellm_core_utils.logging_utils import (
convert_litellm_response_object_to_str,
)
from typing import List
from litellm.llms.custom_httpx.http_handler import (
get_async_httpx_client,
httpxSpecialProvider,
)
import json
from litellm.types.guardrails import GuardrailEventHooks
litellm.set_verbose = True
GUARDRAIL_NAME = "aporia"
class AporiaGuardrail(CustomGuardrail):
def __init__(
self, api_key: Optional[str] = None, api_base: Optional[str] = None, **kwargs
):
self.async_handler = get_async_httpx_client(
llm_provider=httpxSpecialProvider.GuardrailCallback
)
self.aporia_api_key = api_key or os.environ["APORIO_API_KEY"]
self.aporia_api_base = api_base or os.environ["APORIO_API_BASE"]
super().__init__(**kwargs)
#### CALL HOOKS - proxy only ####
def transform_messages(self, messages: List[dict]) -> List[dict]:
supported_openai_roles = ["system", "user", "assistant"]
default_role = "other" # for unsupported roles - e.g. tool
new_messages = []
for m in messages:
if m.get("role", "") in supported_openai_roles:
new_messages.append(m)
else:
new_messages.append(
{
"role": default_role,
**{key: value for key, value in m.items() if key != "role"},
}
)
return new_messages
async def prepare_aporia_request(
self, new_messages: List[dict], response_string: Optional[str] = None
) -> dict:
data: dict[str, Any] = {}
if new_messages is not None:
data["messages"] = new_messages
if response_string is not None:
data["response"] = response_string
# Set validation target
if new_messages and response_string:
data["validation_target"] = "both"
elif new_messages:
data["validation_target"] = "prompt"
elif response_string:
data["validation_target"] = "response"
verbose_proxy_logger.debug("Aporia AI request: %s", data)
return data
async def make_aporia_api_request(
self, new_messages: List[dict], response_string: Optional[str] = None
):
data = await self.prepare_aporia_request(
new_messages=new_messages, response_string=response_string
)
_json_data = json.dumps(data)
"""
export APORIO_API_KEY=<your key>
curl https://gr-prd-trial.aporia.com/some-id \
-X POST \
-H "X-APORIA-API-KEY: $APORIO_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"messages": [
{
"role": "user",
"content": "This is a test prompt"
}
],
}
'
"""
response = await self.async_handler.post(
url=self.aporia_api_base + "/validate",
data=_json_data,
headers={
"X-APORIA-API-KEY": self.aporia_api_key,
"Content-Type": "application/json",
},
)
verbose_proxy_logger.debug("Aporia AI response: %s", response.text)
if response.status_code == 200:
# check if the response was flagged
_json_response = response.json()
action: str = _json_response.get(
"action"
) # possible values are modify, passthrough, block, rephrase
if action == "block":
raise HTTPException(
status_code=400,
detail={
"error": "Violated guardrail policy",
"aporia_ai_response": _json_response,
},
)
async def async_post_call_success_hook(
self,
data: dict,
user_api_key_dict: UserAPIKeyAuth,
response,
):
from litellm.proxy.common_utils.callback_utils import (
add_guardrail_to_applied_guardrails_header,
)
"""
Use this for the post call moderation with Guardrails
"""
event_type: GuardrailEventHooks = GuardrailEventHooks.post_call
if self.should_run_guardrail(data=data, event_type=event_type) is not True:
return
response_str: Optional[str] = convert_litellm_response_object_to_str(response)
if response_str is not None:
await self.make_aporia_api_request(
response_string=response_str, new_messages=data.get("messages", [])
)
add_guardrail_to_applied_guardrails_header(
request_data=data, guardrail_name=self.guardrail_name
)
pass
async def async_moderation_hook( ### 👈 KEY CHANGE ###
self,
data: dict,
user_api_key_dict: UserAPIKeyAuth,
call_type: Literal[
"completion",
"embeddings",
"image_generation",
"moderation",
"audio_transcription",
],
):
from litellm.proxy.common_utils.callback_utils import (
add_guardrail_to_applied_guardrails_header,
)
event_type: GuardrailEventHooks = GuardrailEventHooks.during_call
if self.should_run_guardrail(data=data, event_type=event_type) is not True:
return
# old implementation - backwards compatibility
if (
await should_proceed_based_on_metadata(
data=data,
guardrail_name=GUARDRAIL_NAME,
)
is False
):
return
new_messages: Optional[List[dict]] = None
if "messages" in data and isinstance(data["messages"], list):
new_messages = self.transform_messages(messages=data["messages"])
if new_messages is not None:
await self.make_aporia_api_request(new_messages=new_messages)
add_guardrail_to_applied_guardrails_header(
request_data=data, guardrail_name=self.guardrail_name
)
else:
verbose_proxy_logger.warning(
"Aporia AI: not running guardrail. No messages in data"
)
pass
|