Skip to main content

aimon.Detect

Class: Detect​

A simple class based decorator that can be used for real-time evaluations or for continuous monitoring for detecting quality issues in LLM-generated text.

Constructor​

Detect(values_returned, api_key=None, config=None, async_mode=False, publish=False, application_name=None, model_name=None)

Parameters:​

  • values_returned (list): A list of values in the order returned by the decorated function. Acceptable values are 'generated_text', 'context', 'user_query', 'instructions'.
  • api_key (str, optional): The API key to use for the AIMon client. If not provided, it will be read from the 'AIMON_API_KEY' environment variable.
  • config (dict, optional): A dictionary of configuration options for the detector. Default is {'hallucination': {'detector_name': 'default'}}.
  • async_mode (bool, optional): If True, the detect() function will return immediately with a DetectResult object. Default is False.
  • publish (bool, optional): If True, the payload will be published to AIMon and can be viewed on the AIMon UI. Default is False.
  • application_name (str, optional): Required if publish is True. The name of the application in AIMon.
  • model_name (str, optional): Required if publish is True. The name of the model in AIMon.

Raises:​

  • ValueError: If API key is None, values_returned is empty or doesn't contain 'context', or if publish is True and either application_name or model_name is not provided.

Methods​

__call__(func)​

This method is called when the Detect instance is used as a decorator. It wraps the decorated function and handles the detection process.

Parameters:​
  • func (callable): The function to be decorated.
Returns:​
  • A wrapped version of the input function that includes AIMon detection.

Usage Example​

from aimon import Detect
import os

detect = Detect(
    values_returned=['context', 'generated_text', 'user_query'],
    api_key=os.getenv('AIMON_API_KEY'),
    config={
        'hallucination': {'detector_name': 'default'},
        'toxicity': {'detector_name': 'default'}
    },
    publish=True,
    application_name='my_summarization_app',
    model_name='gpt-3.5-turbo'
)

def your_llm_function(context, query):
    # your LLM function implementation here
    return f"Summary of '{context}' based on query: {query}"

@detect
def generate_summary(context, query):
    summary = your_llm_function(context, query)
    return context, summary, query

context = "The quick brown fox jumps over the lazy dog."
query = "Summarize the given text."
context, summary, query, aimon_result = generate_summary(context, query)

print(f"Hallucination score: {aimon_result.detect_response.hallucination['score']}")
print(f"Toxicity score: {aimon_result.detect_response.toxicity['score']}")

Notes​

  • The values_returned list must contain 'context' and should include 'generated_text'.
  • If async_mode is True, publish is automatically set to True.
  • When publish is True, both application_name and model_name must be provided.