Response Consistency is the measure of how well the generated response aligns with both the question asked and the context provided.

In evaluating response consistency, it is important to assess whether the information provided in the response directly addresses the query posed by the user and is coherent with any additional context given.

Columns required:

  • question: The question asked by the user
  • context: Information retrieved to answer the question
  • response: The response given by the model

​
How to use it?

from uptrain import EvalLLM, Evals

OPENAI_API_KEY = "sk-********************"  # Insert your OpenAI key here

data = [{
    "question": "How is pneumonia treated?",
    "context": "Pneumonia is an infection that inflames the air sacs in one or both lungs. It is typically treated with antibiotics, rest, and supportive care. The choice of antibiotics depends on the type of pneumonia and its severity."
    "response": "Pneumonia can be treated with over-the-counter painkillers, and rest is not necessary for recovery."
}]

eval_llm = EvalLLM(openai_api_key=OPENAI_API_KEY)

res = eval_llm.evaluate(
    data = data,
    checks = [Evals.RESPONSE_CONSISTENCY]
)
By default, we are using GPT 3.5 Turbo for evaluations. If you want to use a different model, check out this tutorial.

Sample Response:

[
   {
      "score_response_consistency": 0.0,
      "explanation_response_consistency": " \"The given LLM response doesn't answer the user query at all because it fails to provide any information on how pneumonia is actually treated. The response only mentions over-the-counter painkillers and dismisses the need for rest, which is not a comprehensive or accurate representation of pneumonia treatment. The user will be highly dissatisfied with this answer as it lacks crucial information on antibiotics, hospitalization, oxygen therapy, and other essential treatments for pneumonia.\"\n"
   }
]
A higher response consistency score reflects that the generated response aligns with both the question asked and the context provided.

The response states the use of painkillers for the treatment of pneumonia.

This contradicts the context which states that pneumonia is typically treated with antibiotics, rest, and supportive care

Resulting in a low response consistency score.

​
How it works?

We evaluate response consistency through the following steps:

  • Generating an argument as to why the given response is appropriate for the question asked.
  • Rating the generated argument on a score of 0 to 1, as per how logical the argument seems to be.

Tutorial

Open this tutorial in GitHub

Have Questions?

Join our community for any questions or requests