Judge
Base class for building custom evaluation scorers.
Subclass Judge and implement the score method to create your own
evaluation logic.
class ContainsAnswer extends Judge<BinaryResponse> {
async score(data: Example): Promise<BinaryResponse> {
const expected = (data.get("expected_output") as string).toLowerCase();
const actual = (data.get("actual_output") as string).toLowerCase();
return {
value: actual.includes(expected),
reason: actual.includes(expected) ? "Found" : "Not found",
};
}
}