Add longer accuracy test on CI (#1049)
This commit is contained in:
68
test/srt/test_eval_accuracy_large.py
Normal file
68
test/srt/test_eval_accuracy_large.py
Normal file
@@ -0,0 +1,68 @@
|
||||
import unittest
|
||||
from types import SimpleNamespace
|
||||
|
||||
from sglang.srt.utils import kill_child_process
|
||||
from sglang.test.run_eval import run_eval
|
||||
from sglang.test.test_utils import (
|
||||
DEFAULT_MODEL_NAME_FOR_TEST,
|
||||
DEFAULT_URL_FOR_TEST,
|
||||
popen_launch_server,
|
||||
)
|
||||
|
||||
|
||||
class TestEvalAccuracyLarge(unittest.TestCase):
|
||||
|
||||
@classmethod
|
||||
def setUpClass(cls):
|
||||
cls.model = DEFAULT_MODEL_NAME_FOR_TEST
|
||||
cls.base_url = "http://127.0.0.1:7157"
|
||||
cls.process = popen_launch_server(
|
||||
cls.model,
|
||||
cls.base_url,
|
||||
timeout=300,
|
||||
other_args=["--log-level-http", "warning"],
|
||||
)
|
||||
|
||||
@classmethod
|
||||
def tearDownClass(cls):
|
||||
kill_child_process(cls.process.pid)
|
||||
|
||||
def test_mmlu(self):
|
||||
args = SimpleNamespace(
|
||||
base_url=self.base_url,
|
||||
model=self.model,
|
||||
eval_name="mmlu",
|
||||
num_examples=None,
|
||||
num_threads=2048,
|
||||
)
|
||||
|
||||
metrics = run_eval(args)
|
||||
assert metrics["score"] >= 0.70
|
||||
|
||||
def test_human_eval(self):
|
||||
args = SimpleNamespace(
|
||||
base_url=self.base_url,
|
||||
model=self.model,
|
||||
eval_name="humaneval",
|
||||
num_examples=None,
|
||||
num_threads=2048,
|
||||
)
|
||||
|
||||
metrics = run_eval(args)
|
||||
assert metrics["score"] >= 0.65
|
||||
|
||||
def test_mgsm_en(self):
|
||||
args = SimpleNamespace(
|
||||
base_url=self.base_url,
|
||||
model=self.model,
|
||||
eval_name="mgsm_en",
|
||||
num_examples=None,
|
||||
num_threads=2048,
|
||||
)
|
||||
|
||||
metrics = run_eval(args)
|
||||
assert metrics["score"] >= 0.85
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
unittest.main()
|
||||
Reference in New Issue
Block a user