tests/test_report.py

from korbyte.benchmark import _flatten_strings
from korbyte.report import _aggregate


def test_aggregate_uses_weighted_counts() -> None:
    metrics = {
        "one": {
            "tokens": 10,
            "characters": 30,
            "utf8_bytes": 50,
            "whitespace_units": 5,
            "elapsed_s_median": 1.0,
        },
        "two": {
            "tokens": 20,
            "characters": 60,
            "utf8_bytes": 100,
            "whitespace_units": 10,
            "elapsed_s_median": 2.0,
        },
    }
    aggregate = _aggregate(metrics)
    assert aggregate["tokens"] == 30
    assert aggregate["characters_per_token"] == 3.0
    assert aggregate["bytes_per_token"] == 5.0
    assert aggregate["fertility"] == 2.0


def test_dialogue_extraction_uses_turn_text_only() -> None:
    dialogue = [
        {"role": "user", "text": "안녕하세요.", "state": ["상태"]},
        {"role": "sys", "text": "무엇을 도와드릴까요?", "state": []},
    ]
    assert list(_flatten_strings(dialogue)) == ["안녕하세요.", "무엇을 도와드릴까요?"]