Initial Commit

Dockerfile

@@ -0,0 +1,13 @@
+FROM python:3.10.13
+
+WORKDIR /app
+
+COPY requirements.txt requirements.txt
+RUN pip install -r requirements.txt
+
+COPY . .
+
+ENV GR_SERVER_NAME=0.0.0.0
+EXPOSE 7860
+
+CMD ["python", "app.py"]

README.md

@@ -1,4 +1,3 @@
----
 title: Demo Chat Gpt
 emoji: 💻
 colorFrom: indigo
@@ -8,6 +7,3 @@ sdk_version: 4.44.0
 app_file: app.py
 pinned: false
 license: apache-2.0
----
-
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference

app.py

@@ -0,0 +1,33 @@
+import os
+import gradio as gr
+import instrumentation
+
+instrumentation.init("llm-chat-app")
+
+from langchain.globals import set_debug
+from langchain_core.prompts import ChatPromptTemplate
+from langchain_core.output_parsers import StrOutputParser
+from langchain_openai import ChatOpenAI
+
+set_debug(True)
+import logging
+
+logging.basicConfig(level=logging.DEBUG)
+llm = ChatOpenAI(temperature=0.5, max_tokens=100, model="gpt-3.5-turbo")
+
+output_parser = StrOutputParser()
+prompt = ChatPromptTemplate.from_messages([
+        ("system", "You are Responsible AI assistant to the user. "),
+        ("user", "{input}")
+    ])
+
+def handle_message(message, _history):
+    chain = prompt | llm | output_parser
+    return chain.invoke({"input": message})
+
+server_name = os.environ.get("GR_SERVER_NAME", "127.0.0.1")
+server_port = os.environ.get("GR_SERVER_PORT", "7860")
+
+options = {}
+
+gr.ChatInterface(handle_message, **options).launch(server_name=server_name, server_port=int(server_port))

docker-compose.yml

@@ -0,0 +1,12 @@
+version: '3.3'
+services:
+  llm-chat-app:
+    build: ./
+    ports:
+      - 7860:7860
+    environment:
+      - OPENAI_BASE_URL=${OPENAI_BASE_URL}
+      - OPENAI_API_KEY=${OPENAI_API_KEY}
+      - TRACELOOP_BASE_URL=${TRACELOOP_BASE_URL}
+      - TRACELOOP_API_KEY=${TRACELOOP_API_KEY}
+

instrumentation.py

@@ -0,0 +1,5 @@
+from traceloop.sdk import Traceloop
+
+# TRACELOOP_BASE_URL=https://ingestor:8080
+def init(app_name: str):
+    Traceloop.init(app_name, disable_batch=True)

pyproject.toml

@@ -0,0 +1,17 @@
+[tool.poetry]
+name = "llm-chat-app"
+version = "0.1.0"
+description = ""
+authors = ["jchauhan <[email protected]>"]
+readme = "README.md"
+
+[tool.poetry.dependencies]
+python = "^3.11"
+gradio = "^4.43.0"
+datasets = "^2.21.0"
+traceloop-sdk = "^0.30.0"
+
+
+[build-system]
+requires = ["poetry-core"]
+build-backend = "poetry.core.masonry.api"

requirements.txt

@@ -0,0 +1,128 @@
+aiofiles==23.2.1
+aiohttp==3.9.3
+aiosignal==1.3.1
+altair==5.2.0
+annotated-types==0.6.0
+anyio==4.2.0
+async-timeout==4.0.3
+attrs==23.2.0
+backoff==2.2.1
+certifi==2024.2.2
+charset-normalizer==3.3.2
+click==8.1.7
+colorama==0.4.6
+contourpy==1.2.0
+cycler==0.12.1
+dataclasses-json==0.6.4
+Deprecated==1.2.14
+distro==1.9.0
+exceptiongroup==1.2.0
+fastapi==0.109.2
+ffmpy==0.3.1
+filelock==3.13.1
+fonttools==4.48.1
+frozenlist==1.4.1
+fsspec==2024.2.0
+googleapis-common-protos==1.62.0
+gradio==4.17.0
+gradio_client==0.9.0
+grpcio==1.60.1
+h11==0.14.0
+httpcore==1.0.2
+httpx==0.26.0
+huggingface-hub==0.20.3
+idna==3.6
+importlib-metadata==6.11.0
+importlib-resources==6.1.1
+inflection==0.5.1
+Jinja2==3.1.3
+jsonpatch==1.33
+jsonpointer==2.4
+jsonschema==4.21.1
+jsonschema-specifications==2023.12.1
+kiwisolver==1.4.5
+langchain==0.1.6
+langchain-community==0.0.19
+langchain-core==0.1.22
+langchain-openai==0.0.5
+langsmith==0.0.87
+markdown-it-py==3.0.0
+MarkupSafe==2.1.5
+marshmallow==3.20.2
+matplotlib==3.8.2
+mdurl==0.1.2
+monotonic==1.6
+multidict==6.0.5
+mypy-extensions==1.0.0
+numpy==1.26.4
+openai==1.12.0
+opentelemetry-api==1.22.0
+opentelemetry-exporter-otlp-proto-common==1.22.0
+opentelemetry-exporter-otlp-proto-grpc==1.22.0
+opentelemetry-exporter-otlp-proto-http==1.22.0
+opentelemetry-instrumentation==0.43b0
+opentelemetry-instrumentation-anthropic==0.10.4
+opentelemetry-instrumentation-bedrock==0.10.4
+opentelemetry-instrumentation-chromadb==0.10.4
+opentelemetry-instrumentation-cohere==0.10.4
+opentelemetry-instrumentation-dbapi==0.43b0
+opentelemetry-instrumentation-haystack==0.10.4
+opentelemetry-instrumentation-langchain==0.10.4
+opentelemetry-instrumentation-llamaindex==0.10.4
+opentelemetry-instrumentation-openai==0.10.4
+opentelemetry-instrumentation-pinecone==0.10.4
+opentelemetry-instrumentation-pymysql==0.43b0
+opentelemetry-instrumentation-replicate==0.10.4
+opentelemetry-instrumentation-requests==0.43b0
+opentelemetry-instrumentation-transformers==0.10.4
+opentelemetry-instrumentation-urllib3==0.43b0
+opentelemetry-instrumentation-vertexai==0.10.4
+opentelemetry-instrumentation-watsonx==0.10.4
+opentelemetry-proto==1.22.0
+opentelemetry-sdk==1.22.0
+opentelemetry-semantic-conventions==0.43b0
+opentelemetry-semantic-conventions-ai==0.0.19
+opentelemetry-util-http==0.43b0
+orjson==3.9.13
+packaging==23.2
+pandas==2.2.0
+pillow==10.2.0
+posthog==3.4.0
+protobuf==4.25.2
+pydantic==2.6.1
+pydantic_core==2.16.2
+pydub==0.25.1
+Pygments==2.17.2
+pyparsing==3.1.1
+python-dateutil==2.8.2
+python-multipart==0.0.7
+pytz==2024.1
+PyYAML==6.0.1
+referencing==0.33.0
+regex==2023.12.25
+requests==2.31.0
+rich==13.7.0
+rpds-py==0.17.1
+ruff==0.2.1
+semantic-version==2.10.0
+shellingham==1.5.4
+six==1.16.0
+sniffio==1.3.0
+SQLAlchemy==2.0.25
+starlette==0.36.3
+tenacity==8.2.3
+tiktoken==0.5.2
+tomlkit==0.12.0
+toolz==0.12.1
+tqdm==4.66.1
+traceloop-sdk==0.10.4
+typer==0.9.0
+typing-inspect==0.9.0
+typing_extensions==4.9.0
+tzdata==2023.4
+urllib3==2.2.0
+uvicorn==0.27.0.post1
+websockets==11.0.3
+wrapt==1.16.0
+yarl==1.9.4
+zipp==3.17.0

tests/scripts/send_attack_prompts_2chatapp.py

@@ -0,0 +1,83 @@
+import sys
+import os
+import random
+from time import sleep
+from gradio_client import Client
+
+current_dir = os.path.dirname(os.path.abspath(__file__))
+sys.path.append(os.path.dirname(current_dir))
+
+from datasets import load_dataset
+
+def send_message(client, message):
+    _apps = ["fintechgpt", "healthgpt", "mydoc", "knowledge-centre", "assistantgpt"]
+
+    # Compute exponential weights for service names
+    factor = 2  # Control the steepness of the exponential decrease
+    app_weights = [factor ** -i for i in range(len(_apps))]
+
+    # Select a service_name randomly based on the exponential weights
+    service_name = random.choices(_apps, weights=app_weights, k=1)[0]
+
+    # Define providers and corresponding models with weights
+    providers_models = {
+        "Meta AI": (["LLaMA 65B", "LLaMA 33B"], 10),
+        "Mistral": (["Mistral 7B"], 8),
+        "Anthropic": (["Claude 3.5 Sonnet"], 15),
+        "OpenAI": (["GPT-3", "GPT-3.5", "GPT-4", "GPT-4o"], 25),
+        "Google": (["Gemini Ultra", "Gemini Pro", "Gemini Nano", "Lamda", "Palm"], 20),
+        "Databricks": (["Dolly"], 5),
+        "IBM AI": (["Watson NLP"], 5),
+        "Azure AI": (["Azure OpenAI", "Custom GPT-3.5"], 7),
+        "Snowflake": (["Snowflake GPT"], 3),
+        "Krutrim": (["Krutrim LLM"], 2),
+        "Baidu": (["Ernie 4.0"], 10),
+        "Stability AI": (["StableLM 3B", "StableLM 7B"], 4),
+        "Microsoft": (["Orca", "Phi-1"], 8)
+    }
+
+    # Extract providers and their weights
+    providers = list(providers_models.keys())
+    provider_weights = [providers_models[provider][1] for provider in providers]
+
+    # Randomly select a provider based on weights
+    selected_provider = random.choices(providers, weights=provider_weights, k=1)[0]
+
+    # Randomly select a model from the selected provider
+    selected_model = random.choice(providers_models[selected_provider][0])
+
+    result = client.predict(
+                message,	# str  in 'Message' Textbox component
+                api_name="/chat"
+            )
+    print(f"Sending message: {message} \n\t Result {result}")    
+    
+
+def main():
+    prompt_injection_dataset = load_dataset("deepset/prompt-injections")
+    toxicity_prompts = load_dataset("allenai/real-toxicity-prompts")
+
+    nats_url = os.environ.get('CHAT_APP_URL', "http://localhost:7860/")
+    client = Client(nats_url)
+
+    while True:
+        # Collect messages
+        prompt_injection_sample = prompt_injection_dataset["train"].shuffle(seed=42).select(range(10))
+        toxicity_sample = toxicity_prompts["train"].shuffle(seed=42).select(range(10))
+        plain_messages = ["this is test conversation" for _ in range(10)]
+
+        # Combine all messages into a single list
+        all_messages = [msg["text"] for msg in prompt_injection_sample] + \
+                       [msg["prompt"]["text"] for msg in toxicity_sample] + \
+                       plain_messages
+
+        # Shuffle the combined list to mix message types
+        random.shuffle(all_messages)
+
+        # Send each message
+        for message in all_messages:
+            send_message(client, message)
+            sleep(random.uniform(0.5, 2))  # Random sleep between 0.5 to 2 seconds
+
+if __name__ == "__main__":
+    main()

tests/scripts/simulate_attack_prompts.py

@@ -0,0 +1,286 @@
+import json
+import random
+import time
+from datetime import datetime
+import os
+import requests
+from datasets import load_dataset
+from time import sleep
+
+
+TEMPLATE_JSON=json.loads("""
+{
+    "resource_spans": [
+      {
+        "scope_spans": [
+          {
+            "spans": [
+              {
+                "trace_id": "NQ01459b3A+aAHE+JwGWNQ==",
+                "end_time_unix_nano": "1725721375827041000",
+                "span_id": "0PCGpTowmVo=",
+                "kind": "SPAN_KIND_CLIENT",
+                "name": "openai.chat",
+                "start_time_unix_nano": "1725721375188928425",
+                "attributes": [
+                  {
+                    "value": {
+                      "string_value": "chat"
+                    },
+                    "key": "llm.request.type"
+                  },
+                  {
+                    "value": {
+                      "string_value": "OpenAI"
+                    },
+                    "key": "llm.vendor"
+                  },
+                  {
+                    "value": {
+                      "string_value": "gpt-3.5-turbo"
+                    },
+                    "key": "llm.request.model"
+                  },
+                  {
+                    "value": {
+                      "int_value": "100"
+                    },
+                    "key": "llm.request.max_tokens"
+                  },
+                  {
+                    "value": {
+                      "double_value": 0.5
+                    },
+                    "key": "llm.temperature"
+                  },
+                  {
+                    "value": {
+                      "string_value": "None"
+                    },
+                    "key": "llm.headers"
+                  },
+                  {
+                    "value": {
+                      "string_value": "system"
+                    },
+                    "key": "llm.prompts.0.role"
+                  },
+                  {
+                    "value": {
+                      "string_value": "You are Responsible AI assistant to the user. "
+                    },
+                    "key": "llm.prompts.0.content"
+                  },
+                  {
+                    "value": {
+                      "string_value": "user"
+                    },
+                    "key": "llm.prompts.1.role"
+                  },
+                  {
+                    "value": {
+                      "string_value": "hello this my test message"
+                    },
+                    "key": "llm.prompts.1.content"
+                  },
+                  {
+                    "value": {
+                      "string_value": "gpt-3.5-turbo-0125"
+                    },
+                    "key": "llm.response.model"
+                  },
+                  {
+                    "value": {
+                      "int_value": "35"
+                    },
+                    "key": "llm.usage.total_tokens"
+                  },
+                  {
+                    "value": {
+                      "int_value": "9"
+                    },
+                    "key": "llm.usage.completion_tokens"
+                  },
+                  {
+                    "value": {
+                      "int_value": "26"
+                    },
+                    "key": "llm.usage.prompt_tokens"
+                  },
+                  {
+                    "value": {
+                      "string_value": "stop"
+                    },
+                    "key": "llm.completions.0.finish_reason"
+                  },
+                  {
+                    "value": {
+                      "string_value": "assistant"
+                    },
+                    "key": "llm.completions.0.role"
+                  },
+                  {
+                    "value": {
+                      "string_value": "Hello! How can I assist you today?"
+                    },
+                    "key": "llm.completions.0.content"
+                  }
+                ],
+                "status": {}
+              }
+            ],
+            "scope": {
+              "name": "opentelemetry.instrumentation.openai.v1",
+              "version": "0.10.4"
+            }
+          }
+        ],
+        "resource": {
+          "attributes": [
+            {
+              "value": {
+                "string_value": "llm-chat-app"
+              },
+              "key": "service.name"
+            }
+          ]
+        }
+      }
+    ]
+}
+
+""")
+
+
+def generate_random_id(size):
+    return ''.join(random.choices('ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789+/=', k=size))
+
+
+def generate_json_from_template(template, provider, model, service_name, message):    
+    # Generate random trace_id and span_id
+    trace_id = generate_random_id(22)
+    span_id = generate_random_id(12)
+    
+    # Get current time in nanoseconds
+    current_time_ns = int(time.time() * 1e9)
+    
+    # Update trace_id, span_id, times, provider, model, and service_name
+    template['resource_spans'][0]['scope_spans'][0]['spans'][0]['trace_id'] = trace_id
+    template['resource_spans'][0]['scope_spans'][0]['spans'][0]['span_id'] = span_id
+    template['resource_spans'][0]['scope_spans'][0]['spans'][0]['start_time_unix_nano'] = str(current_time_ns)
+    template['resource_spans'][0]['scope_spans'][0]['spans'][0]['end_time_unix_nano'] = str(current_time_ns + random.randint(100000000, 1000000000))  # Random duration
+
+    # Update provider, model, and service_name information in attributes
+    for attr in template['resource_spans'][0]['scope_spans'][0]['spans'][0]['attributes']:
+        if attr['key'] == 'llm.vendor':
+            attr['value']['string_value'] = provider
+        elif attr['key'] == 'llm.request.model':
+            attr['value']['string_value'] = model
+        elif attr['key'] == 'llm.prompts.1.content':  # Update user message content
+            attr['value']['string_value'] = message
+
+    # Update service_name in the resource attributes
+    for attr in template['resource_spans'][0]['resource']['attributes']:
+        if attr['key'] == 'service.name':
+            attr['value']['string_value'] = service_name
+
+    # Return the modified JSON
+    return json.dumps(template)
+
+
+def send_json_to_remote(json_data):
+    # Get environment variables
+    base_url = os.getenv('TRACELOOP_BASE_URL')
+    api_key = os.getenv('TRACELOOP_API_KEY')
+    
+    if not base_url or not api_key:
+        raise EnvironmentError("TRACELOOP_BASE_URL or TRACELOOP_API_KEY is not set in environment variables.")
+
+    # Set the headers and URL
+    url = f"{base_url}/v1/traces"
+    headers = {
+        'Content-Type': 'application/json',
+        'Authorization': f"Bearer {api_key}"
+    }
+
+    # Send the POST request
+    response = requests.post(url, headers=headers, data=json_data)
+
+    # Check the response status
+    if response.status_code == 200:
+        print("Data successfully sent!")
+    else:
+        print(f"Failed to send data. Status Code: {response.status_code}, Response: {response.text}")
+    return (response.status_code, response.text)
+
+def send_message(message):
+    _apps = ["fintechgpt", "healthgpt", "mydoc", "knowledge-centre", "assistantgpt"]
+
+    # Compute exponential weights for service names
+    factor = 2  # Control the steepness of the exponential decrease
+    app_weights = [factor ** -i for i in range(len(_apps))]
+
+    # Select a service_name randomly based on the exponential weights
+    service_name = random.choices(_apps, weights=app_weights, k=1)[0]
+
+    # Define providers and corresponding models with weights
+    providers_models = {
+        "Meta AI": (["LLaMA 65B", "LLaMA 33B"], 10),
+        "Mistral": (["Mistral 7B"], 8),
+        "Anthropic": (["Claude 3.5 Sonnet"], 15),
+        "OpenAI": (["GPT-3", "GPT-3.5", "GPT-4", "GPT-4o"], 25),
+        "Google": (["Gemini Ultra", "Gemini Pro", "Gemini Nano", "Lamda", "Palm"], 20),
+        "Databricks": (["Dolly"], 5),
+        "IBM AI": (["Watson NLP"], 5),
+        "Azure AI": (["Azure OpenAI", "Custom GPT-3.5"], 7),
+        "Snowflake": (["Snowflake GPT"], 3),
+        "Krutrim": (["Krutrim LLM"], 2),
+        "Baidu": (["Ernie 4.0"], 10),
+        "Stability AI": (["StableLM 3B", "StableLM 7B"], 4),
+        "Microsoft": (["Orca", "Phi-1"], 8)
+    }
+
+    # Extract providers and their weights
+    providers = list(providers_models.keys())
+    provider_weights = [providers_models[provider][1] for provider in providers]
+
+    # Randomly select a provider based on weights
+    selected_provider = random.choices(providers, weights=provider_weights, k=1)[0]
+
+    # Randomly select a model from the selected provider
+    selected_model = random.choice(providers_models[selected_provider][0])
+
+    # Generate the JSON with the selected provider, model, and service_name
+    output_json = generate_json_from_template(TEMPLATE_JSON,  selected_provider, selected_model, service_name, message)
+
+    # Send the JSON to the remote server
+    return send_json_to_remote(output_json)
+
+
+def main():
+    prompt_injection_dataset = load_dataset("deepset/prompt-injections")
+    toxicity_prompts = load_dataset("allenai/real-toxicity-prompts")
+
+    for i in range(2):
+        # Collect messages
+        prompt_injection_sample = prompt_injection_dataset["train"].shuffle(seed=42).select(range(10))
+        toxicity_sample = toxicity_prompts["train"].shuffle(seed=42).select(range(10))
+        plain_messages = ["this is test conversation" for _ in range(10)]
+
+        # Combine all messages into a single list
+        all_messages = [msg["text"] for msg in prompt_injection_sample] + \
+                       [msg["prompt"]["text"] for msg in toxicity_sample] + \
+                       plain_messages
+
+        # Shuffle the combined list to mix message types
+        random.shuffle(all_messages)
+
+        # Send each message
+        for message in all_messages:
+            print(f"Sending Message {message}")
+            send_message(message)
+            sleep(random.uniform(2, 4))  # Random sleep between 0.5 to 2 seconds
+
+if __name__ == "__main__":
+    main()
+