File size: 4,196 Bytes
9269c10
 
b850722
 
 
 
9269c10
b850722
 
af688eb
 
b850722
 
 
9269c10
 
 
 
 
 
af688eb
 
 
 
 
 
 
 
 
 
2946856
 
9269c10
 
 
 
 
 
 
 
 
 
 
 
 
c32f628
9269c10
 
 
 
 
 
c32f628
9269c10
 
 
2946856
9269c10
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2946856
af688eb
 
b20052a
af688eb
9269c10
 
af688eb
9269c10
af688eb
 
b20052a
af688eb
 
b20052a
af688eb
 
 
 
 
6d0856c
2946856
af688eb
 
2946856
 
af688eb
 
 
 
 
 
 
 
 
2946856
af688eb
 
 
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
import importlib

import streamlit as st
import weave
from dotenv import load_dotenv

from guardrails_genie.guardrails import GuardrailManager
from guardrails_genie.llm import OpenAIModel

st.title(":material/robot: Guardrails Genie Playground")

load_dotenv()
weave.init(project_name="guardrails-genie")

if "guardrails" not in st.session_state:
    st.session_state.guardrails = []
if "guardrail_names" not in st.session_state:
    st.session_state.guardrail_names = []
if "guardrails_manager" not in st.session_state:
    st.session_state.guardrails_manager = None
if "initialize_guardrails" not in st.session_state:
    st.session_state.initialize_guardrails = False
if "system_prompt" not in st.session_state:
    st.session_state.system_prompt = ""
if "user_prompt" not in st.session_state:
    st.session_state.user_prompt = ""
if "test_guardrails" not in st.session_state:
    st.session_state.test_guardrails = False
if "llm_model" not in st.session_state:
    st.session_state.llm_model = None


def initialize_guardrails():
    st.session_state.guardrails = []
    for guardrail_name in st.session_state.guardrail_names:
        if guardrail_name == "PromptInjectionSurveyGuardrail":
            survey_guardrail_model = st.sidebar.selectbox(
                "Survey Guardrail LLM", ["", "gpt-4o-mini", "gpt-4o"]
            )
            if survey_guardrail_model:
                st.session_state.guardrails.append(
                    getattr(
                        importlib.import_module("guardrails_genie.guardrails"),
                        guardrail_name,
                    )(llm_model=OpenAIModel(model_name=survey_guardrail_model))
                )
        else:
            st.session_state.guardrails.append(
                getattr(
                    importlib.import_module("guardrails_genie.guardrails"),
                    guardrail_name,
                )()
            )
    st.session_state.guardrails_manager = GuardrailManager(
        guardrails=st.session_state.guardrails
    )


openai_model = st.sidebar.selectbox(
    "OpenAI LLM for Chat", ["", "gpt-4o-mini", "gpt-4o"]
)
chat_condition = openai_model != ""

guardrails = []

guardrail_names = st.sidebar.multiselect(
    label="Select Guardrails",
    options=[
        cls_name
        for cls_name, cls_obj in vars(
            importlib.import_module("guardrails_genie.guardrails")
        ).items()
        if isinstance(cls_obj, type) and cls_name != "GuardrailManager"
    ],
)
st.session_state.guardrail_names = guardrail_names

if st.sidebar.button("Initialize Guardrails") and chat_condition:
    st.session_state.initialize_guardrails = True

if st.session_state.initialize_guardrails:
    with st.sidebar.status("Initializing Guardrails..."):
        initialize_guardrails()
        st.session_state.llm_model = OpenAIModel(model_name=openai_model)

    user_prompt = st.text_area("User Prompt", value="")
    st.session_state.user_prompt = user_prompt

    test_guardrails_button = st.button("Test Guardrails")
    st.session_state.test_guardrails = test_guardrails_button

    if st.session_state.test_guardrails:
        with st.sidebar.status("Running Guardrails..."):
            guardrails_response, call = st.session_state.guardrails_manager.guard.call(
                st.session_state.guardrails_manager, prompt=st.session_state.user_prompt
            )

        if guardrails_response["safe"]:
            st.markdown(
                f"\n\n---\nPrompt is safe! Explore prompt trace on [Weave]({call.ui_url})\n\n---\n"
            )

            with st.sidebar.status("Generating response from LLM..."):
                response, call = st.session_state.llm_model.predict.call(
                    st.session_state.llm_model,
                    user_prompts=st.session_state.user_prompt,
                )
            st.markdown(
                response.choices[0].message.content
                + f"\n\n---\nExplore LLM generation trace on [Weave]({call.ui_url})"
            )
        else:
            st.warning("Prompt is not safe!")
            st.markdown(guardrails_response["summary"])
            st.markdown(f"Explore prompt trace on [Weave]({call.ui_url})")