Spaces:
Running
Running
geekyrakshit
commited on
Commit
·
c89e6e0
1
Parent(s):
159baa9
update: app
Browse files
.gitignore
CHANGED
@@ -167,4 +167,5 @@ test.py
|
|
167 |
temp.txt
|
168 |
**.csv
|
169 |
binary-classifier/
|
170 |
-
wandb/
|
|
|
|
167 |
temp.txt
|
168 |
**.csv
|
169 |
binary-classifier/
|
170 |
+
wandb/
|
171 |
+
artifacts/
|
application_pages/chat_app.py
CHANGED
@@ -1,4 +1,5 @@
|
|
1 |
import importlib
|
|
|
2 |
|
3 |
import streamlit as st
|
4 |
import weave
|
@@ -7,27 +8,27 @@ from dotenv import load_dotenv
|
|
7 |
from guardrails_genie.guardrails import GuardrailManager
|
8 |
from guardrails_genie.llm import OpenAIModel
|
9 |
|
10 |
-
st.title(":material/robot: Guardrails Genie Playground")
|
11 |
-
|
12 |
-
load_dotenv()
|
13 |
-
weave.init(project_name="guardrails-genie")
|
14 |
|
15 |
-
|
16 |
-
|
17 |
-
|
18 |
-
|
19 |
-
if "
|
20 |
-
|
21 |
-
if "
|
22 |
-
|
23 |
-
if "
|
24 |
-
|
25 |
-
if "
|
26 |
-
|
27 |
-
if "
|
28 |
-
|
29 |
-
if "
|
30 |
-
|
|
|
|
|
|
|
|
|
31 |
|
32 |
|
33 |
def initialize_guardrails():
|
@@ -44,18 +45,30 @@ def initialize_guardrails():
|
|
44 |
guardrail_name,
|
45 |
)(llm_model=OpenAIModel(model_name=survey_guardrail_model))
|
46 |
)
|
47 |
-
|
48 |
-
st.
|
49 |
-
|
50 |
-
|
51 |
-
|
52 |
-
|
|
|
|
|
53 |
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
54 |
st.session_state.guardrails_manager = GuardrailManager(
|
55 |
guardrails=st.session_state.guardrails
|
56 |
)
|
57 |
|
58 |
|
|
|
|
|
|
|
59 |
openai_model = st.sidebar.selectbox(
|
60 |
"OpenAI LLM for Chat", ["", "gpt-4o-mini", "gpt-4o"]
|
61 |
)
|
@@ -97,7 +110,7 @@ if st.session_state.initialize_guardrails:
|
|
97 |
|
98 |
if guardrails_response["safe"]:
|
99 |
st.markdown(
|
100 |
-
f"\n\n---\nPrompt is safe! Explore
|
101 |
)
|
102 |
|
103 |
with st.sidebar.status("Generating response from LLM..."):
|
|
|
1 |
import importlib
|
2 |
+
import os
|
3 |
|
4 |
import streamlit as st
|
5 |
import weave
|
|
|
8 |
from guardrails_genie.guardrails import GuardrailManager
|
9 |
from guardrails_genie.llm import OpenAIModel
|
10 |
|
|
|
|
|
|
|
|
|
11 |
|
12 |
+
def initialize_session_state():
|
13 |
+
load_dotenv()
|
14 |
+
weave.init(project_name=os.getenv("WEAVE_PROJECT"))
|
15 |
+
|
16 |
+
if "guardrails" not in st.session_state:
|
17 |
+
st.session_state.guardrails = []
|
18 |
+
if "guardrail_names" not in st.session_state:
|
19 |
+
st.session_state.guardrail_names = []
|
20 |
+
if "guardrails_manager" not in st.session_state:
|
21 |
+
st.session_state.guardrails_manager = None
|
22 |
+
if "initialize_guardrails" not in st.session_state:
|
23 |
+
st.session_state.initialize_guardrails = False
|
24 |
+
if "system_prompt" not in st.session_state:
|
25 |
+
st.session_state.system_prompt = ""
|
26 |
+
if "user_prompt" not in st.session_state:
|
27 |
+
st.session_state.user_prompt = ""
|
28 |
+
if "test_guardrails" not in st.session_state:
|
29 |
+
st.session_state.test_guardrails = False
|
30 |
+
if "llm_model" not in st.session_state:
|
31 |
+
st.session_state.llm_model = None
|
32 |
|
33 |
|
34 |
def initialize_guardrails():
|
|
|
45 |
guardrail_name,
|
46 |
)(llm_model=OpenAIModel(model_name=survey_guardrail_model))
|
47 |
)
|
48 |
+
elif guardrail_name == "PromptInjectionClassifierGuardrail":
|
49 |
+
classifier_model_name = st.sidebar.selectbox(
|
50 |
+
"Classifier Guardrail Model",
|
51 |
+
[
|
52 |
+
"",
|
53 |
+
"ProtectAI/deberta-v3-base-prompt-injection-v2",
|
54 |
+
"wandb://geekyrakshit/guardrails-genie/model-6rwqup9b:v3",
|
55 |
+
],
|
56 |
)
|
57 |
+
if classifier_model_name != "":
|
58 |
+
st.session_state.guardrails.append(
|
59 |
+
getattr(
|
60 |
+
importlib.import_module("guardrails_genie.guardrails"),
|
61 |
+
guardrail_name,
|
62 |
+
)(model_name=classifier_model_name)
|
63 |
+
)
|
64 |
st.session_state.guardrails_manager = GuardrailManager(
|
65 |
guardrails=st.session_state.guardrails
|
66 |
)
|
67 |
|
68 |
|
69 |
+
initialize_session_state()
|
70 |
+
st.title(":material/robot: Guardrails Genie Playground")
|
71 |
+
|
72 |
openai_model = st.sidebar.selectbox(
|
73 |
"OpenAI LLM for Chat", ["", "gpt-4o-mini", "gpt-4o"]
|
74 |
)
|
|
|
110 |
|
111 |
if guardrails_response["safe"]:
|
112 |
st.markdown(
|
113 |
+
f"\n\n---\nPrompt is safe! Explore guardrail trace on [Weave]({call.ui_url})\n\n---\n"
|
114 |
)
|
115 |
|
116 |
with st.sidebar.status("Generating response from LLM..."):
|
application_pages/evaluation_app.py
CHANGED
@@ -64,10 +64,22 @@ def initialize_guardrail():
|
|
64 |
guardrail_name,
|
65 |
)(llm_model=OpenAIModel(model_name=survey_guardrail_model))
|
66 |
)
|
67 |
-
|
68 |
-
|
69 |
-
|
|
|
|
|
|
|
|
|
|
|
70 |
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
71 |
st.session_state.guardrails = guardrails
|
72 |
st.session_state.guardrail_manager = GuardrailManager(guardrails=guardrails)
|
73 |
|
|
|
64 |
guardrail_name,
|
65 |
)(llm_model=OpenAIModel(model_name=survey_guardrail_model))
|
66 |
)
|
67 |
+
elif guardrail_name == "PromptInjectionClassifierGuardrail":
|
68 |
+
classifier_model_name = st.sidebar.selectbox(
|
69 |
+
"Classifier Guardrail Model",
|
70 |
+
[
|
71 |
+
"",
|
72 |
+
"ProtectAI/deberta-v3-base-prompt-injection-v2",
|
73 |
+
"wandb://geekyrakshit/guardrails-genie/model-6rwqup9b:v3",
|
74 |
+
],
|
75 |
)
|
76 |
+
if classifier_model_name:
|
77 |
+
st.session_state.guardrails.append(
|
78 |
+
getattr(
|
79 |
+
import_module("guardrails_genie.guardrails"),
|
80 |
+
guardrail_name,
|
81 |
+
)(model_name=classifier_model_name)
|
82 |
+
)
|
83 |
st.session_state.guardrails = guardrails
|
84 |
st.session_state.guardrail_manager = GuardrailManager(guardrails=guardrails)
|
85 |
|
guardrails_genie/guardrails/__init__.py
CHANGED
@@ -1,8 +1,11 @@
|
|
1 |
-
from .injection import
|
|
|
|
|
|
|
2 |
from .manager import GuardrailManager
|
3 |
|
4 |
__all__ = [
|
5 |
"PromptInjectionSurveyGuardrail",
|
6 |
-
"
|
7 |
"GuardrailManager",
|
8 |
]
|
|
|
1 |
+
from .injection import (
|
2 |
+
PromptInjectionClassifierGuardrail,
|
3 |
+
PromptInjectionSurveyGuardrail,
|
4 |
+
)
|
5 |
from .manager import GuardrailManager
|
6 |
|
7 |
__all__ = [
|
8 |
"PromptInjectionSurveyGuardrail",
|
9 |
+
"PromptInjectionClassifierGuardrail",
|
10 |
"GuardrailManager",
|
11 |
]
|
guardrails_genie/guardrails/injection/__init__.py
CHANGED
@@ -1,4 +1,4 @@
|
|
1 |
-
from .protectai_guardrail import
|
2 |
from .survey_guardrail import PromptInjectionSurveyGuardrail
|
3 |
|
4 |
-
__all__ = ["PromptInjectionSurveyGuardrail", "
|
|
|
1 |
+
from .protectai_guardrail import PromptInjectionClassifierGuardrail
|
2 |
from .survey_guardrail import PromptInjectionSurveyGuardrail
|
3 |
|
4 |
+
__all__ = ["PromptInjectionSurveyGuardrail", "PromptInjectionClassifierGuardrail"]
|
guardrails_genie/guardrails/injection/protectai_guardrail.py
CHANGED
@@ -5,16 +5,25 @@ import weave
|
|
5 |
from transformers import AutoModelForSequenceClassification, AutoTokenizer, pipeline
|
6 |
from transformers.pipelines.base import Pipeline
|
7 |
|
|
|
|
|
8 |
from ..base import Guardrail
|
9 |
|
10 |
|
11 |
-
class
|
12 |
model_name: str = "ProtectAI/deberta-v3-base-prompt-injection-v2"
|
13 |
_classifier: Optional[Pipeline] = None
|
14 |
|
15 |
def model_post_init(self, __context):
|
16 |
-
|
17 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
18 |
self._classifier = pipeline(
|
19 |
"text-classification",
|
20 |
model=model,
|
|
|
5 |
from transformers import AutoModelForSequenceClassification, AutoTokenizer, pipeline
|
6 |
from transformers.pipelines.base import Pipeline
|
7 |
|
8 |
+
import wandb
|
9 |
+
|
10 |
from ..base import Guardrail
|
11 |
|
12 |
|
13 |
+
class PromptInjectionClassifierGuardrail(Guardrail):
|
14 |
model_name: str = "ProtectAI/deberta-v3-base-prompt-injection-v2"
|
15 |
_classifier: Optional[Pipeline] = None
|
16 |
|
17 |
def model_post_init(self, __context):
|
18 |
+
if self.model_name.startswith("wandb://"):
|
19 |
+
api = wandb.Api()
|
20 |
+
artifact = api.artifact(self.model_name.removeprefix("wandb://"))
|
21 |
+
artifact_dir = artifact.download()
|
22 |
+
tokenizer = AutoTokenizer.from_pretrained(artifact_dir)
|
23 |
+
model = AutoModelForSequenceClassification.from_pretrained(artifact_dir)
|
24 |
+
else:
|
25 |
+
tokenizer = AutoTokenizer.from_pretrained(self.model_name)
|
26 |
+
model = AutoModelForSequenceClassification.from_pretrained(self.model_name)
|
27 |
self._classifier = pipeline(
|
28 |
"text-classification",
|
29 |
model=model,
|