Hallucination

When To Use Hallucination Eval Template

This LLM Eval detects if the output of a model is a hallucination based on contextual data.

This Eval is specifically designed to detect hallucinations in generated answers from private or retrieved data. The Eval detects if an AI answer to a question is a hallucination based on the reference data used to generate the answer.

This Eval is designed to check for hallucinations on private data, specifically on data that is fed into the context window from retrieval.

It is not designed to check hallucinations on what the LLM was trained on. It is not useful for random public fact hallucinations. E.g. “What was Michael Jordan’s birthday?”

Hallucination Eval Template

In this task, you will be presented with a query, some context and a response. The response
is generated to the question based on the context. The response may contain false
information. You must use the context to determine if the response to the question
contains false information, if the response is a hallucination of facts. Your objective is
to determine whether the response text contains factual information and is not a
hallucination. A 'hallucination' refers to a response that is not based on the context or
assumes information that is not available in the context. Your response should be a single
word: either 'factual' or 'hallucinated', and it should not include any other text or
characters. 'hallucinated' indicates that the response provides factually inaccurate
information to the query based on the context. 'factual' indicates that the response to
the question is correct relative to the context, and does not contain made up
information. Please read the query and context carefully before determining your
response.

[BEGIN DATA]
************
[Query]: {input}
************
[Context]: {context}
************
[Response]: {output}
************
[END DATA]

Is the response above factual or hallucinated based on the query and context?

We are continually iterating our templates, view the most up-to-date template on GitHub.

How To Run the Hallucination Eval

The HallucinationEvaluator requires three inputs called input, output, and context. You can use the .describe() method on any evaluator to learn more about it, including it’s input_schema which has information about required inputs.

from phoenix.evals.llm import LLM
from phoenix.evals.metrics import HallucinationEvaluator

# initialize LLM and evaluator
llm = LLM(model="gpt-4o", provider="openai")
hallucination = HallucinationEvaluator(llm=llm)

# use the .describe() method to inspect the input_schema of any evaluator
print(hallucination_evaluator.describe())
>>> {'name': 'hallucination',
 'source': 'llm',
 'direction': 'maximize',
 'input_schema': {'properties': {
   'input': {'description': 'The input query.',
    'title': 'Input',
    'type': 'string'},
   'output': {'description': 'The response to the query.',
    'title': 'Output',
    'type': 'string'},
   'context': {'description': 'The context or reference text.',
    'title': 'Context',
    'type': 'string'}},
  'required': ['input', 'output', 'context'],
  'title': 'HallucinationInputSchema',
  'type': 'object'}}

# let's test on one example
eval_input = {
    "input": "Where is the Eiffel Tower located?",
    "context": "The Eiffel Tower is located in Paris, France. It was constructed in 1889 as the entrance arch to the 1889 World's Fair.",
    "output": "The Eiffel Tower is located in Paris, France.",
}

scores = hallucination.evaluate(eval_input=eval_input)
print(scores[0])
>>> Score(name='hallucination', score=1.0, label='factual', explanation='The response correctly identifies the location of the Eiffel Tower as stated in the context.', metadata={'model': 'gpt-4o'}, source='llm', direction='maximize')

Benchmark Results

This benchmark was obtained using notebook below. It was run using the HaluEval QA Dataset as a ground truth dataset. Each example in the dataset was evaluating using the HALLUCINATION_PROMPT_TEMPLATE above, then the resulting labels were compared against the is_hallucination label in the HaluEval dataset to generate the confusion matrices below.

Google Colab

colab.research.google.com

GPT-4 Results

Scikit GPT-4

Eval	GPT-4
Precision	0.93
Recall	0.72
F1	0.82

Throughput	GPT-4
100 Samples	105 sec