Write your own LLM Eval class

Have you seen our custom evaluation wrappers? They are a great way to quickly create your own evaluation class.

If you want to write your own evaluation class, you can do so by extending the BaseEvaluator class. You can extend the BaseEvaluator class. See this example You need to implement the following methods:

@property
def _model(self):
    # Which model was used for this evaluation
    pass

@property
def name(self):
    return "UpperCamelCaseName"

@property
def display_name(self):
    return "Evaluation Display Name"

@property
def metric_ids(self) -> List[str]:
    return [MetricType.PASSED.value]

@property
def default_function_arguments(self):
    return {}

@property
def required_args(self):
    # expects an array of strings from ["query", "context", "response", "expected_response", "text"]
    pass

@property
def examples(self):
    pass

def is_failure(self) -> Optional[bool]:
    pass


def _evaluate(self, **kwargs) -> EvalResult:
    pass

Contribute !If you write an evaluator that could be useful for others, please consider raising a PR at https://github.com/athina-ai/athina-evals.

Custom Grading Criteria Overview

Logging

Datasets

Evals

GraphQL API

Deprecated

Write your own LLM Eval class