2024-08-01 21:20:17 -07:00
|
|
|
import unittest
|
|
|
|
|
from types import SimpleNamespace
|
|
|
|
|
|
2024-11-28 00:22:39 -08:00
|
|
|
from sglang.srt.utils import kill_process_tree
|
2024-08-01 21:20:17 -07:00
|
|
|
from sglang.test.run_eval import run_eval
|
2024-08-11 18:27:33 -07:00
|
|
|
from sglang.test.test_utils import (
|
|
|
|
|
DEFAULT_MODEL_NAME_FOR_TEST,
|
2024-08-25 19:02:08 -07:00
|
|
|
DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
|
|
|
|
|
DEFAULT_URL_FOR_TEST,
|
2024-08-11 18:27:33 -07:00
|
|
|
popen_launch_server,
|
|
|
|
|
)
|
2024-08-01 21:20:17 -07:00
|
|
|
|
|
|
|
|
|
2024-08-12 02:21:38 -07:00
|
|
|
class TestEvalAccuracyMini(unittest.TestCase):
|
2024-08-01 21:20:17 -07:00
|
|
|
@classmethod
|
|
|
|
|
def setUpClass(cls):
|
2024-08-04 20:51:55 -07:00
|
|
|
cls.model = DEFAULT_MODEL_NAME_FOR_TEST
|
2024-08-25 19:02:08 -07:00
|
|
|
cls.base_url = DEFAULT_URL_FOR_TEST
|
|
|
|
|
cls.process = popen_launch_server(
|
|
|
|
|
cls.model, cls.base_url, timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH
|
|
|
|
|
)
|
2024-08-01 21:20:17 -07:00
|
|
|
|
|
|
|
|
@classmethod
|
|
|
|
|
def tearDownClass(cls):
|
2024-11-28 00:22:39 -08:00
|
|
|
kill_process_tree(cls.process.pid)
|
2024-08-01 21:20:17 -07:00
|
|
|
|
|
|
|
|
def test_mmlu(self):
|
|
|
|
|
args = SimpleNamespace(
|
|
|
|
|
base_url=self.base_url,
|
|
|
|
|
model=self.model,
|
|
|
|
|
eval_name="mmlu",
|
2024-09-09 05:32:41 -07:00
|
|
|
num_examples=64,
|
2024-08-12 02:21:38 -07:00
|
|
|
num_threads=32,
|
2024-10-22 23:20:43 -07:00
|
|
|
temperature=0.1,
|
2024-08-01 21:20:17 -07:00
|
|
|
)
|
|
|
|
|
|
|
|
|
|
metrics = run_eval(args)
|
2024-11-14 01:30:24 -08:00
|
|
|
self.assertGreaterEqual(metrics["score"], 0.65)
|
2024-08-01 21:20:17 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
if __name__ == "__main__":
|
2024-08-10 15:09:03 -07:00
|
|
|
unittest.main()
|