End-to-end observability with built-in security guardrails.
Project description
OpenLLMTelemetry
openllmtelemetry
is an open-source Python library that provides Open Telemetry integration with Large Language Models (LLMs). It is designed to facilitate tracing applications that leverage LLMs and Generative AI, ensuring better observability and monitoring.
Features
- Easy integration with Open Telemetry for LLM applications.
- Real-time tracing and monitoring of LLM-based systems.
- Enhanced safeguards and insights for your LLM applications.
Installation
To install openllmtelemetry
simply use pip:
pip install openllmtelemetry
Usage 🚀
Here's a basic example of how to use OpenLLMTelemetry in your project:
First you need to setup a few environment variables to specify where you want your LLM telemetry to be sent, and make sure you also have any API keys set for interacting with your LLM and for sending the telemetry to WhyLabs
import os
os.environ["WHYLABS_DEFAULT_DATASET_ID"] = "your-model-id" # e.g. model-1
os.environ["WHYLABS_API_KEY"] = "replace-with-your-whylabs-api-key"
After you verify your env variables are set you can now instrument your app by running the following:
import openllmtelemetry
openllmtelemetry.instrument()
This will automatically instrument your calls to LLMs to gather open telemetry traces and send these to WhyLabs.
Integration: OpenAI
Integration with an OpenAI application is straightforward with openllmtelemetry
package.
First, you need to set a few environment variables. This can be done via your container set up or via code.
import os
os.environ["WHYLABS_API_KEY"] = "<your-whylabs-api-key>"
os.environ["WHYLABS_DEFAULT_DATASET_ID"] = "<your-llm-resource-id>"
os.environ["GUARDRAILS_ENDPOINT"] = "<your container endpoint>"
os.environ["GUARDRAILS_API_KEY"] = "internal-secret-for-whylabs-Secure"
Once this is done, all of your OpenAI interactions will be automatically traced. If you have rulesets enabled for blocking in WhyLabs Secure policy, the library will block requests accordingly
from openai import OpenAI
client = OpenAI()
response = client.chat.completions.create(
model="gpt-3.5-turbo",
messages=[
{
"role": "system",
"content": "You are a helpful chatbot. "
},
{
"role": "user",
"content": "Aren't noodles amazing?"
}
],
temperature=0.7,
max_tokens=64,
top_p=1
)
Integration: Amazon Bedrock
One of the nice things about openllmtelemetry
is that a single call to intrument your app can work across various LLM providers, using the same instrument call above, you can also invoke models using the boto3 client's bedrock-runtime and interaction with LLMs such as Titan and you get the same level of telemetry extracted and sent to WhyLabs
Note: you may have to test that your boto3 credentials are working to be able to use the below example For details see boto3 documentation
import json
import boto3
def bedrock_titan(prompt: str):
try:
model_id = 'amazon.titan-text-express-v1'
brt = boto3.client(service_name='bedrock-runtime')
response = brt.invoke_model(body=json.dumps({"inputText": prompt}), modelId=model_id)
response_body = json.loads(response.get("body").read())
except Exception as error:
logger.error(f"A client error occurred:{error}")
return response_body
response = bedrock_titan("What is your name and what is the origin and reason for that name?")
print(response)
Requirements 📋
- Python 3.8 or higher
- opentelemetry-api
- opentelemetry-sdk
Contributing 👐
Contributions are welcome! For major changes, please open an issue first to discuss what you would like to change. Please make sure to update tests as appropriate.
License 📄
OpenLLMTelemetry is licensed under the Apache-2.0 License. See LICENSE for more details.
Contact 📧
For support or any questions, feel free to contact us at support@whylabs.ai.
Documentation
More documentation can be found here on WhyLabs site: https://whylabs.ai/docs/
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
Hashes for openllmtelemetry-0.0.1b16.tar.gz
Algorithm | Hash digest | |
---|---|---|
SHA256 | 1bd79b50bddc52e5217cdecfb24cabba55ef218d09f8c18075d46a06e4700a87 |
|
MD5 | 37c53549b69f0ec8c0a68b80cffefaff |
|
BLAKE2b-256 | 4005149bdfdc1a0364133e1c34ffe392567eddf6e91b47c2e464a25f5951dde7 |
Hashes for openllmtelemetry-0.0.1b16-py3-none-any.whl
Algorithm | Hash digest | |
---|---|---|
SHA256 | f54a01efe2a5fa493908b75021c73a5d5a2f45e4b64374c705baed8277b21438 |
|
MD5 | 87fae73c2e3289aed5b70d30b3595fc5 |
|
BLAKE2b-256 | a4ea510deddb7a53d031924e12a2667c3c512e9576ea559036d0e10f2e210e3a |