File size: 10,193 Bytes
447ebeb
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
import pytest
import asyncio
import aiohttp, openai
from openai import OpenAI, AsyncOpenAI
from typing import Optional, List, Union
import json
import uuid


async def chat_completion(
    session,
    key,
    messages,
    model: Union[str, List] = "gpt-4",
    guardrails: Optional[List] = None,
):
    url = "http://0.0.0.0:4000/chat/completions"
    headers = {
        "Authorization": f"Bearer {key}",
        "Content-Type": "application/json",
    }

    data = {
        "model": model,
        "messages": messages,
    }

    if guardrails is not None:
        data["guardrails"] = guardrails

    print("data=", data)

    async with session.post(url, headers=headers, json=data) as response:
        status = response.status
        response_text = await response.text()

        print(response_text)
        print()

        if status != 200:
            raise Exception(response_text)

        # response headers
        response_headers = dict(response.headers)
        print("response headers=", response_headers)

        return await response.json(), response_headers


async def generate_key(
    session, guardrails: Optional[List] = None, team_id: Optional[str] = None
):
    url = "http://0.0.0.0:4000/key/generate"
    headers = {"Authorization": "Bearer sk-1234", "Content-Type": "application/json"}
    data = {}
    if guardrails:
        data["guardrails"] = guardrails
    if team_id:
        data["team_id"] = team_id

    async with session.post(url, headers=headers, json=data) as response:
        status = response.status
        response_text = await response.text()

        print(response_text)
        print()

        if status != 200:
            raise Exception(f"Request did not return a 200 status code: {status}")

        return await response.json()


@pytest.mark.asyncio
@pytest.mark.skip(reason="Aporia account disabled")
async def test_llm_guard_triggered_safe_request():
    """
    - Tests a request where no content mod is triggered
    - Assert that the guardrails applied are returned in the response headers
    """
    async with aiohttp.ClientSession() as session:
        response, headers = await chat_completion(
            session,
            "sk-1234",
            model="fake-openai-endpoint",
            messages=[{"role": "user", "content": f"Hello what's the weather"}],
            guardrails=[
                "aporia-post-guard",
                "aporia-pre-guard",
            ],
        )
        await asyncio.sleep(3)

        print("response=", response, "response headers", headers)

        assert "x-litellm-applied-guardrails" in headers

        assert (
            headers["x-litellm-applied-guardrails"]
            == "aporia-pre-guard,aporia-post-guard"
        )


@pytest.mark.asyncio
@pytest.mark.skip(reason="Aporia account disabled")
async def test_llm_guard_triggered():
    """
    - Tests a request where no content mod is triggered
    - Assert that the guardrails applied are returned in the response headers
    """
    async with aiohttp.ClientSession() as session:
        try:
            response, headers = await chat_completion(
                session,
                "sk-1234",
                model="fake-openai-endpoint",
                messages=[
                    {"role": "user", "content": f"Hello my name is [email protected]"}
                ],
                guardrails=[
                    "aporia-post-guard",
                    "aporia-pre-guard",
                ],
            )
            pytest.fail("Should have thrown an exception")
        except Exception as e:
            print(e)
            assert "Aporia detected and blocked PII" in str(e)


@pytest.mark.asyncio
async def test_no_llm_guard_triggered():
    """
    - Tests a request where no content mod is triggered
    - Assert that the guardrails applied are returned in the response headers
    """
    async with aiohttp.ClientSession() as session:
        response, headers = await chat_completion(
            session,
            "sk-1234",
            model="fake-openai-endpoint",
            messages=[{"role": "user", "content": f"Hello what's the weather"}],
            guardrails=[],
        )
        await asyncio.sleep(3)

        print("response=", response, "response headers", headers)

        assert "x-litellm-applied-guardrails" not in headers


@pytest.mark.asyncio
async def test_guardrails_with_api_key_controls():
    """
    - Make two API Keys
        - Key 1 with no guardrails
        - Key 2 with guardrails
    - Request to Key 1 -> should be success with no guardrails
    - Request to Key 2 -> should be error since guardrails are triggered
    """
    async with aiohttp.ClientSession() as session:
        key_with_guardrails = await generate_key(
            session=session,
            guardrails=[
                "bedrock-pre-guard",
            ],
        )

        key_with_guardrails = key_with_guardrails["key"]

        key_without_guardrails = await generate_key(session=session, guardrails=None)

        key_without_guardrails = key_without_guardrails["key"]

        # test no guardrails triggered for key without guardrails
        response, headers = await chat_completion(
            session,
            key_without_guardrails,
            model="fake-openai-endpoint",
            messages=[{"role": "user", "content": f"Hello what's the weather"}],
        )
        await asyncio.sleep(3)

        print("response=", response, "response headers", headers)
        assert "x-litellm-applied-guardrails" not in headers

        # test guardrails triggered for key with guardrails
        response, headers = await chat_completion(
            session,
            key_with_guardrails,
            model="fake-openai-endpoint",
            messages=[{"role": "user", "content": f"Hello my name is [email protected]"}],
        )

        assert "x-litellm-applied-guardrails" in headers
        assert headers["x-litellm-applied-guardrails"] == "bedrock-pre-guard"


@pytest.mark.asyncio
async def test_bedrock_guardrail_triggered():
    """
    - Tests a request where our bedrock guardrail should be triggered
    - Assert that the guardrails applied are returned in the response headers
    """
    async with aiohttp.ClientSession() as session:
        try:
            response, headers = await chat_completion(
                session,
                "sk-1234",
                model="fake-openai-endpoint",
                messages=[{"role": "user", "content": "Hello do you like coffee?"}],
                guardrails=["bedrock-pre-guard"],
            )
            pytest.fail("Should have thrown an exception")
        except Exception as e:
            print(e)
            assert "GUARDRAIL_INTERVENED" in str(e)
            assert "Violated guardrail policy" in str(e)


@pytest.mark.asyncio
async def test_custom_guardrail_during_call_triggered():
    """
    - Tests a request where our bedrock guardrail should be triggered
    - Assert that the guardrails applied are returned in the response headers
    """
    async with aiohttp.ClientSession() as session:
        try:
            response, headers = await chat_completion(
                session,
                "sk-1234",
                model="fake-openai-endpoint",
                messages=[{"role": "user", "content": f"Hello do you like litellm?"}],
                guardrails=["custom-during-guard"],
            )
            pytest.fail("Should have thrown an exception")
        except Exception as e:
            print(e)
            assert "Guardrail failed words - `litellm` detected" in str(e)


async def create_team(session, guardrails: Optional[List] = None):
    url = "http://0.0.0.0:4000/team/new"
    headers = {"Authorization": "Bearer sk-1234", "Content-Type": "application/json"}
    data = {"guardrails": guardrails}

    print("request data=", data)

    async with session.post(url, headers=headers, json=data) as response:
        status = response.status
        response_text = await response.text()

        print(response_text)
        print()

        if status != 200:
            raise Exception(f"Request did not return a 200 status code: {status}")

        return await response.json()


@pytest.mark.asyncio
async def test_guardrails_with_team_controls():
    """
    - Create a team with guardrails
    - Make two API Keys
        - Key 1 not associated with team
        - Key 2 associated with team (inherits team guardrails)
    - Request with Key 1 -> should be success with no guardrails
    - Request with Key 2 -> should error since team guardrails are triggered
    """
    async with aiohttp.ClientSession() as session:

        # Create team with guardrails
        team = await create_team(
            session=session,
            guardrails=[
                "bedrock-pre-guard",
            ],
        )

        print("team=", team)

        team_id = team["team_id"]

        # Create key with team association
        key_with_team = await generate_key(session=session, team_id=team_id)
        key_with_team = key_with_team["key"]

        # Create key without team
        key_without_team = await generate_key(
            session=session,
        )
        key_without_team = key_without_team["key"]

        # Test no guardrails triggered for key without a team
        response, headers = await chat_completion(
            session,
            key_without_team,
            model="fake-openai-endpoint",
            messages=[{"role": "user", "content": "Hello my name is [email protected]"}],
        )
        await asyncio.sleep(3)

        print("response=", response, "response headers", headers)
        assert "x-litellm-applied-guardrails" not in headers

        response, headers = await chat_completion(
            session,
            key_with_team,
            model="fake-openai-endpoint",
            messages=[{"role": "user", "content": "Hello my name is [email protected]"}],
        )

        print("response headers=", json.dumps(headers, indent=4))

        assert "x-litellm-applied-guardrails" in headers
        assert headers["x-litellm-applied-guardrails"] == "bedrock-pre-guard"